1: \documentclass[11pt]{article}

2: 

3: \textwidth=6.5in

4: \textheight=8.9in

5: 

6: \oddsidemargin=0cm

7: \topmargin=-1cm

8: 

9: \usepackage[all]{xy}

10: 

11: \usepackage{mathrsfs} %

12: \usepackage{bbm}

13: \usepackage{enumerate} %

14: 

15: \input{commands.tex}

16: 

17: \begin{document}

18: \setcounter{table}{1}

19: 

20: \begin{center}

21: {

22: \Large\bf

23: 

24: Jordan algebras, exceptional groups,\\[2mm]

25: and higher composition laws

26: }

27: 

28: \vspace{\baselineskip}

29: 

30: {\large Sergei Krutelevich}

31: 

32: \vspace{\baselineskip}

33: 

34: \end{center}

35: 

36: \begin{abstract}

37: 

38: We consider an integral version of the Freudenthal construction relating

39: Jordan algebras and exceptional algebraic groups. We show how this

40: construction is related to higher composition laws of M.~Bhargava in

41: number theory \cite{bh1}.

42: 

43: We propose an algorithmic approach to studying orbit spaces of

44: groups underlying higher composition laws.

45: Using this method we discover two new

46: examples of spaces sharing similar properties, and indicate

47: several more examples of spaces where such composition laws may be

48: introduced.

49: \end{abstract}

50: 

51: \tableofcontents

52: 

53: 

54: \section{Introduction}

55: \subsection{Integral representations of exceptional groups}

56: \label{i:fc}

57: 

58: It is a well-known fact in number theory that there is a one-to-one

59: correspondence between

60: the set of $\SL_2(\zet)$-equivalence classes of integral binary quadratic

61: forms and the set of (narrow) ideal classes in quadratic rings. About two

62: hundred years ago Gauss discovered the law of composition of binary

63: quadratic forms, which turns

64: the set of equivalence classes of primitive forms of a given discriminant

65: $D$ into a group isomorphic to the ideal class group of the quadratic ring

66: of discriminant $D$. This correspondence is a very important tool

67: for doing computations in the ideal class group.

68: 

69: A few years ago M.~Bhargava discovered several more examples of the same

70: kind, which he referred to as higher composition laws. More precisely, he

71: showed that there are other examples of linear groups $G_\zet$

72: and their integral representations $V_\zet$ such that $G_\zet$-orbits in $V_\zet$

73: are in one-to-one correspondence with ideal classes in the rings of

74: integers in number fields, see \cite{bh-phd, bhau, bh1}.

75: M.~Bhargava also noted a surprising connection between spaces underlying higher

76: composition and exceptional Lie groups.

77: 

78: In the present paper we investigate this connection with exceptional

79: groups. We show that there is a natural construction which assigns

80: a cubic Jordan algebra to every space underlying higher composition laws

81: associated with quadratic rings. We study the appropriate orbit spaces using an

82: algorithmic procedure reminiscent of the Gaussian elimination algorithm

83: for the usual integral matrices. This approach allows us to provide two

84: new examples of spaces sharing similar properties, and indicate

85: several more examples of spaces where such composition laws may be

86: introduced.

87: 

88: \medskip

89: Our interest in the study of integral representations of exceptional groups was

90: originally motivated by a question on a standard form of a charge vector in a certain

91: physical model, which appeared in a paper on BPS black holes in string theory by

92: H.~Maldacena, G.~Moore, A.~Strominger \cite{mms}.

93: This question is equivalent to the question on a normal form of a vector

94: in the $27$-dimensional integral representation of the split group of type

95: $E_6$.

96: Such a representation may be constructed via the exceptional Jordan

97: algebra of $3\times 3$ Hermitian matrices over the algebra of octonions.

98: The answer, generalizing the classical results on the Smith normal

99: form for regular integral matrices, was given in our earlier paper

100: \cite{k}, see also Theorem~$\ref{np-orbits}$ below.

101: 

102: It was suggested by G.~Moore that another interesting space to look

103: at in this context is the $56$-dimensional integral representation of the

104: split group of type $E_7$. This representation may be constructed using the

105: $27$-dimensional exceptional Jordan algebra. This procedure is known

106: as the Freudenthal construction \cite{b, fr14},

107: and it can be applied to other cubic Jordan algebras.

108: Given a cubic Jordan algebra~$\J$, the Freudenthal  construction

109: produces a semisimple algebraic group and an irreducible module over it.

110: We will denote the group obtained by $\gr$ and the module by $\mj$

111: (see Subsection~\ref{ss:fc} for details).

112: Moreover, this module is equipped with a quartic form and a symplectic

113: form, which are invariant under the action of the group.

114: When $\J$ is the $27$-dimensional exceptional Jordan

115: algebra, the group $\gr$ is $E_7$ and $\mj$ is its $56$-dimensional

116: representation.

117: 

118: A natural

119: way to construct a Jordan algebra with a cubic form is to consider the

120: space of $3\times 3$-Hermitian matrices over a composition algebra

121: over a field $F$, the cubic form being the determinant of such matrices

122: (see Example~$\ref{ex:jcubic}$ for details). When $F$ is algebraically

123: closed, there are four composition algebras, and they produce four cubic

124: Jordan algebras of dimension $6, 9, 15, 27$.

125: 

126: The application of the Freudenthal construction yields a module of

127: dimension $14, 20, 32, 56$ for a certain simple algebraic group

128: of type $C_3, A_5, D_6, E_7$.

129: This module is in fact a prehomogeneous vector space for this

130: algebraic group, and it is natural to study orbits under the action of

131: the group.

132: In the case $F=\com$, the classification of orbits of the one-dimensional

133: subspaces\footnote{We avoid

134: using the term ``projective" here, since it is used in completely

135: different sense in the rest of the paper.}

136: arising this way was obtained by J.-L.~Clerc~\cite{clerc}.

137: 

138: Lie algebras of the algebraic groups of type $C_3, A_5, D_6, E_7$ appear

139: in the third row of the Freudenthal-Tits magic square. The groups,

140: produced by the Freudenthal construction, also

141: appear in the last row of the

142: ``magic triangle" of subgroups associated to the exceptional

143: series of P.~Deligne, see \cite{dg}. In addition, the representation $\m$

144: of the group $\gr$ is the {\em preferred} representation of this group

145: in the sense of \cite{dg}.

146: 

147: In the present paper we consider an integral version of the Freudenthal

148: construction.

149: In this case the quartic form (the {\em norm}) takes on integer values.

150: We study integral forms of the split groups of type $A_5, D_6, E_7$

151: determined by the Freudenthal construction, and

152: we obtain the following result

153: on the structure of integral orbits in the $\zet$-modules of

154: dimension $20, 32, 56$.

155: \medskip

156: 

157: \hspace{-\parindent}{\bf Theorem~$\ref{thm:main}$}

158: {\em

159: Let $(G_\zet, \mz)$ be one the following pairs

160: $$

161: \Bigl(\SL_6(\zet), \wedge^3(\zet^6)\Bigr),\quad

162: \Bigl(D_6(\zet), \mbox{\rm half-spin}_\zet\Bigr),\quad

163: \Bigl(E_7(\zet), V(\omega_7)_\zet\Bigr),

164: $$

165: Then

166: \begin{itemize}

167: 

168: \item

169: The $G_\zet$-invariant quartic form (the norm) on the module $\mz$

170: has values congruent to $0$ or $1\, (\!\mod 4)$.

171: 

172: \item

173: Let $n$ be an integer $\equiv 0$ or $1\, (\!\mod 4)$.

174: The group $G_\zet$ acts

175: transitively on the set of {\em projective} elements of norm $n$.

176: 

177: \item

178: If $n$ is a fundamental discriminant\footnote{

179: An integer $n$ is called a fundamental discriminant if $n$ is squarefree

180: and $\equiv 1 (\mod 4)$

181: or $n=4k$, where $k$ is a squarefree integer that is $\equiv 2$ or $3 (\mod 4)$.

182: The result stated in the theorem applies in the case $n=1$.

183: },

184: then every element of norm $n$ is projective, and hence in this case

185: $G_\zet$ acts transitively on the set of elements of norm $n$.

186: \end{itemize}

187: }

188: 

189: 

190: The assertion of this theorem for $\SL_6(\zet)$-orbits

191: on $\wedge^3(\zet^6)$ (when $n\ne 0$) was proved by M.~Bhargava in

192: \cite[Theorem 7]{bh1}, using the correspondence with (narrow) ideal

193: classes in quadratic orders.

194: 

195: Our approach, based on the Freudenthal construction, allows us to treat

196: all values of the norm (including $n=0$) uniformly.

197: Our statement of Theorem~$\ref{thm:main}$

198: was motivated by \cite{bh1}, but the proof

199: presented here is completely independent.

200: It is algorithmic in nature,

201: and may be thought of as a more sophisticated version of

202: the Gaussian algorithm bringing an integer matrix to the Smith

203: normal form by elementary row and column transformations.

204: The extended version of Theorem~$\ref{thm:main}$ for the degenerate orbits

205: (corresponding to the case $n=0$)

206: is given in Theorem~$\ref{thm2}$.

207: 

208: 

209: The concept of a projective element was introduced in \cite{bh1}.

210: The idea is that these elements are mapped to invertible ideal classes under

211: Bhargava's correspondence. In the case of $\wedge^3(\zet^6)$ they were

212: defined via their $\SL_6(\zet)$ orbit representatives

213: (cf. Definition~$\ref{def:proja}$(a,b)).

214: It follows from our considerations that projective elements have a very convenient

215: description in terms of the partial derivatives of the quartic invariant

216: of the module (see Corollary~$\ref{cor:nabla}$). This assertion is proved for the spaces

217: $\mz$ as

218: in Theorem~$\ref{thm:main}$, but it remains valid in other spaces associated to ideal

219: classes in quadratic orders. (see Subsection~\ref{i:hcl}).

220: 

221: %

222: We also indicate a link between the Freudenthal construction and the

223: original Gauss's composition of quadratic forms in the Appendix at the end

224: of the paper.

225: 

226: \medskip

227: As a by-product of our considerations we obtain the classification of

228: orbits in the four ``Freudenthal" modules in the case of a field. Similar

229: results were obtained in \cite{clerc} and \cite[Section 5.3]{lm}

230: for orbits of one-dimensional subspaces in the case of complex numbers.

231: Our

232: techniques however are purely algebraic, which allows us to extend those

233: results to orbits of {\em elements} over arbitrary field of char$\ne 2,3$.

234: 

235: \medskip

236: \hspace{-\parindent}{\bf Theorem~$\ref{thm1}$}

237: {\em

238: \begin{enumerate}

239: \item[\rm (i)]

240: {

241: Let $F$ be a field of char$\ne 2,3$, let $\calg$ be the split composition algebra

242: $\bin, \quat, \oct$ of dimension $2,4,8$ over $F$, and let $\J=\h\calg$.

243: Let $(G, \m)$ be the pair (group, module) produced from $\J$

244: by the Freudenthal construction.

245: 

246: Then

247: 

248: \begin{itemize}

249: \item

250: There exists a $G$-invariant quartic form (the norm) on the module $\m$.

251: 

252: \item

253: The group $G$ acts

254: transitively on the sets of elements of rank $1,2$, and $3$ in the module

255: $\m$.

256: 

257: \item

258: In the case of rank\/ $4$ the group $G$ acts transitively on the set of

259: elements of a given norm $k$, for any $k\in F$, $k\ne 0$.

260: \end{itemize}

261: All these orbits are distinct, and the union of these orbits and $\{0\}$

262: is the whole module~$\m$.

263: }

264: 

265: \item[\rm (ii)]

266: If in addition every element of $F$ is a square,

267: then the same results apply to the pair $(G, \m)$ obtained

268: from the Jordan algebra $\J=\h F$.

269: \end{enumerate}

270: This construction yields the classification of orbits

271: of the irreducible representations

272: of simple algebraic groups listed in the following table

273: $$

274: \begin{array}{|c|c|c|}

275: \hline

276: \qquad \J \qquad       &  \mbox{Type of }G & \mbox{Highest weight of }\ \m\\

277: \hline

278: \h{\F}    &  C_3 & \omega_3\\

279: \h{\bin}  &  A_5 & \omega_3\\

280: \h{\quat} &  D_6 & \omega_5\ {\rm or}\ \omega_6\\

281: \h{\oct}  &  E_7 & \omega_7\\

282: \hline

283: \end{array}

284: $$

285: }

286: 

287: 

288: 

289: 

290: 

291: \subsection{The Freudenthal construction and higher composition laws}

292: \label{i:hcl}

293: 

294: M.~Bhargava showed that Gauss's composition law

295: is one in a series of at least 14 examples of the same kind (higher

296: composition laws) \cite{bh-phd, bhau, bh1}.

297: There is a certain integral linear group $G_\zet$ and

298: a module $V_\zet$ over it in each of his examples

299: such that $G_\zet$-orbits in $V_\zet$ can be described

300: in terms of ideal classes of orders in a number field.

301: Spaces underlying higher composition laws are closely related to

302: prehomogeneous vector spaces classified by M.~Sato and T.~Kimura \cite{SK}. In

303: particular, each of them is equipped with a polynomial, which is invariant

304: under the action of the appropriate group.

305: 

306: An examination of the table of higher composition laws \cite[Table 1]{bhau}

307: shows that each

308: space associated to a quadratic ring (except Gauss's composition) has a

309: polynomial invariant of degree four. A more detailed analysis suggests

310: that for each pair $(G_\zet, \ V_\zet)$ associated to a quadratic ring,

311: there exists a cubic Jordan algebra $\J_\zet$, such that

312: $(G_\zet, \ V_\zet)$ is essentially the pair produced by the Freudenthal

313: construction. These observations are summarized in Table 1 below.

314: 

315: %

316: \bigskip

317: \caption{The Freudenthal construction and higher composition laws}

318: \nopagebreak

319: $$

320: \begin{array}{|c|c|c|c|c|c|c|}

321: \hline

322: \# &

323: \J &

324: %

325: \phantom{\Bigl(}

326: {\dim}\ \J&

327: %

328: {\rm Group}\ \gr &

329: {\rm Rep.\ \m(\J)} &

330: {\dim}\ \m(\J)  &

331: (\lie, \gamma) \\

332: \hline

333: \hline

334: 

335: 

336: 1 & F\phantom{\Bigl(}   & 1  & \SL_2 & {\rm Sym}^3 V_2     & 4  & G_2, \alpha_2 \\

337: 2 & F\oplus F   & 2  & (\SL_2)^2 & V_2\tensor {\rm Sym}^2 V_2     & 6  & B_3, \alpha_2 \\

338: 3 & \h{0}=F\oplus F\oplus F\phantom{\Bigl(}

339: & 3 & (\SL_2)^3& V_2\otimes V_2\otimes V_2 & 8  & D_4, \alpha_2 \\

340: %

341: 4 & F\oplus Q_4& 5 & \SL_2\times \SL_{4} & V_2\tensor \wedge^2 V_4     & 12  & D_5, \alpha_2\\

342: 5 & \h\bin  & 9  & \SL_6 & V(\omega _3)     & 20  & E_6, \alpha_2 \\

343: \hline

344: \hline

345: 6 & \h\quat & 15 & D_6 & \mbox{half-spin} & 32  & E_7, \alpha_1  \\

346: 7 & \h\oct  & 27 & E_7 & {\rm minuscule}     & 56  & E_8, \alpha_8  \\

347: \hline

348: \hline

349: 8 & F\oplus Q_n, n\ge 3 & 1+n & {\rm SL_2\times SO}_{n+2} & V_2\tensor V(\omega_1)

350: & 2n+4  &{\mathfrak s\mathfrak o}_{n+6}, \alpha_2  \\

351: %

352: 9 & \h{F}   & 6  & C_3 & V(\omega _3)     & 14  & F_4, \alpha_1 \\

353: \hline

354: \end{array}

355: $$

356: %

357: 

358: \renewcommand{\abstractname}{Summary of Table 1}

359: \nopagebreak

360: \begin{abstract}

361: {\bf Headers:} the table lists a semisimple

362: Jordan algebra $\J$ with a cubic form, its

363: dimension, the group $\gr$ (up to a finite subgroup or finite covering)

364: and its module $\mj$ produced by the Freudenthal

365: construction, and the dimension of $\mj$.

366: The last column lists a certain simple Lie

367: algebra $\lie$ and its simple root $\gamma$ associated to $\gr, \mj$ (see

368: subsection \ref{lieconn}).

369: 

370: {\bf Notation:} in the second column, $\h{\calg}$ stands for the Jordan

371: algebra of $3\times 3$-Hermitian matrices over the composition algebra

372: $C$; and $Q_n$ stands for the simple Jordan algebra of a quadratic form

373: (details are found in Subsection~$\ref{ss:ex}$). $V_n$ denotes the standard $n$-dimensional

374: module for $\SL_n$, and $V(\omega_i)$ denotes the simple module with

375: highest weight $\omega_i$ over the appropriate group.

376: Note that rows $2,3,4$ are special cases of row $8$ with $n=1,2,4$.

377: 

378: \medskip

379: 

380: There is a natural integral structure in each Jordan algebra $\J$ above. It

381: induces an integral structure in $\mj$ and $\gr$.

382: The first five rows of the table contain the integral group

383: and a module, which appear in the list of

384: higher composition laws

385: (see \cite[Table 1]{bhau} or \cite{bh1}).

386: 

387: Rows $6$ and $7$ do not appear in the list of higher composition laws \cite{bhau}.

388: However, Theorem~$\ref{thm:main}$  implies that the structure of orbits

389: of the projective elements is

390: the same as in row $\#5$. These are the first examples

391: of groups, more complex than a direct product of $\SL$'s,

392: acting in the spaces underlying higher composition laws.

393: 

394: Finally, the Freudenthal construction produces two more examples (rows $8$

395: and $9$). We conjecture that for $\#8$ the orbit structure can be

396: described in terms of known examples of small dimension. And the row $9$

397: could possibly produce a new example of a space with a composition law.

398: \end{abstract}

399: 

400: It would be interesting to describe orbits of the group $D_6(\zet)$ in the

401: $32$-dimensional module and of the group $E_7(\zet)$ in the

402: $56$-dimensional module in terms of the ideal classes of the appropriate

403: rings in a way similar to how this is done in \cite{bh1} for $\SL_6(\zet)$

404: orbits in $\wedge^3 \zet^6$.

405: 

406: Finally, we note that our algorithmic approach allows one to develop,

407: at least in some cases, a reduction theory similar to that for binary

408: quadratic forms.

409: 

410: \subsection{Connection with Lie algebras}

411: \label{lieconn}

412: 

413: Another interesting feature of the spaces underlying higher composition

414: laws is their remarkable connection with exceptional groups.

415: M. Bhargava showed that each of his pairs (group, space) may be described

416: in terms of the Levi decomposition of parabolic subgroups

417: of exceptional groups.

418: 

419: This connection can be stated more precisely for the Freudenthal

420: construction, and hence for orbit spaces associated to quadratic rings.

421: Here we present a more algebraic (as opposed to Lie group) realization of

422: this construction.

423: Namely, we consider a simple Lie algebra $\lie$ with a $5$-grading associated

424: to the minimal nilpotent orbit in $\lie$. From this setup it

425: is possible to extract an algebraic group $G$ and a $G$-module $M$ with

426: $G$-invariant quartic and skew-symmetric bilinear form, such that $(G,M)$

427: will produce all pairs arising from the Freudenthal construction.

428: 

429: The detailed description of this procedure is given below. Most of the

430: material of this subsection has appeared earlier elsewhere. We used

431: \cite[Section~2]{gw} and \cite{clerc} as the references.

432: For simplicity we assume in this subsection that the ground field $F$ is

433: the field of complex numbers $\com$.

434: 

435: \medskip

436: Let $\lie$ be a complex simple Lie algebra, and we assume that $\lie\ne

437: A_n, C_n$.

438: Let $\Phi$ be its root system with respect to a Cartan subalgebra

439: $\cart$; let $\Phi^+$ be a set of positive roots,

440: and let $\Delta$ be its collection of simple roots.

441: Let $\beta$ be the

442: highest root of $\Phi ^+$. We normalize the inner product

443: $\langle\cdot,\cdot\rangle$ on the real span of roots by requiring

444: $\langle\beta,\beta\rangle=2$. Then it follows that for any

445: $\alpha\in\Phi^+$, $\langle\alpha,\beta\rangle=0,1,2$; and

446: $\langle\alpha,\beta\rangle=2$ iff $\alpha=\beta$.

447: 

448: 

449: We consider the grading

450: \begin{equation}\label{grading}

451: \lie=\lie_{-2}\oplus \lie_{-1}\oplus\lie_{0}\oplus\lie_{1}\oplus\lie_{2},

452: \end{equation}

453: 

454: given by

455: $$

456: \lie_0=\cart\oplus\bigoplus_{\stackrel{\alpha\in\Phi}

457: {\langle\alpha,\beta\rangle=0}}

458: \lie_\alpha,\qquad

459: \lie_k=\bigoplus_{\stackrel{\alpha\in\Phi}

460: {\langle\alpha,\beta\rangle=k}}

461: \lie_\alpha\quad

462: k=\pm 1, \pm 2.

463: $$

464: 

465: In particular, we have $\lie_2=\lie_\beta,\

466: \lie_{-2}=\lie_{\beta}$.

467: 

468: We will also need to consider the extended Dynkin diagram of $\lie$

469: (with the additional vertex corresponding to the root $-\beta$).

470: We let $\gamma$ denote the unique simple

471: root whose vertex is connected to the

472: extended vertex $-\beta$. The standard references for root systems

473: and Dynkin diagrams is \cite{b1}.

474: The appropriate diagrams are also provided in~\cite{gw}.

475: 

476: The grading (\ref{grading}) has the following property:

477: the root space  $\lie_\alpha$ is contained in $\lie_k$

478: if and only if

479: the decomposition of $\alpha$ in the basis of simple roots $\Delta$

480: has coefficient $k$ at the root $\gamma$

481: $(k=0,\pm 1, \pm 2)$.

482: 

483: There exists a subalgebra $\ml\subset\lie_0$

484: such that $\lie_0=[\lie_{-2}, \lie_2]\oplus \ml$.

485: $\ml$ is a semisimple Lie

486: algebra, and its Dynkin diagram can be obtained from the Dynkin diagram of

487: $\lie$ by removing the vertex corresponding to the simple root~$\gamma$.

488: 

489: We let $M_\com$ be a complex connected Lie group, whose Lie algebra is

490: $\ml$. The group $M_\com$ and the Lie algebra $\ml$ act on the space

491: $\lie_1$. The resulting pairs $(M_\com, \lie_1)$ are tabulated in

492: \cite[Table~2.6]{gw}. An examination of that table shows that when

493: $\lie\ne A_n, C_n$, pairs $(M_\com, \lie_1)$ are the same (up to a finite

494: covering) as the pairs $(\gr, \m)$ arising from the Freudenthal construction

495: and listed in Table~1 above. In particular, the last column of that table

496: lists the simple Lie algebra $\lie$ and the simple root $\gamma$ such that

497: the ``Freudenthal" pair $(\gr,\m)$ is obtained from the construction

498: described above.

499: 

500: \medskip

501: Next, we are going show how one can describe the ``Freudenthal" quartic and

502: symplectic forms on $\lie_1\ (=\m)$ invariant with respect to $M_\com$.

503: 

504: The Lie bracket on $\lie$ induces a map $\wedge^2 \lie_1\to \lie_\beta$

505: which is $\lie_0$-equivariant. Since $\lie_\beta$ is one-dimensional, this

506: map defines a symplectic bilinear form $\{\cdot, \cdot\}$ on $\lie_1$.

507: In addition, we have $[\ml, \lie_2]=0$, and this implies that

508: $\{\cdot, \cdot\}$ is invariant with respect to $\ml$ (and hence $M_\com$).

509: 

510: Finally, for $X\in \lie_1$ we consider the map

511: $$

512: ({\adlie\,}X)^4: \ \lie_{-2}\ \longrightarrow\ \lie_2.

513: $$

514: 

515: For fixed root vectors $x_{\pm \beta}\in \lie_{\pm 2}$ we have

516: \begin{equation}\label{qmap}

517: ({\adlie\,}X)^4: \ x_{-\beta}\ \mto \ P(X)\,x_\beta \qquad

518: \mbox{for some }P(X)\in\com.

519: \end{equation}

520: 

521: The function $P:\lie_1\to\com$ defined by (\ref{qmap}) is in fact non-zero

522: homogeneous

523: polynomial function of degree four, invariant with respect to $\ml$ and

524: $M_\com$ \cite[Proposition~3.1]{clerc}. Since $\ml$ acts irreducibly on

525: $\lie_1$, we can normalize the polynomial $P$ so that it coincides with

526: the Freudenthal quartic form.

527: 

528: \medskip

529: With a little more work one can even extract the cubic Jordan algebra $\J$

530: from $\ml$. Namely, in each such $\ml$ there will be a simple root

531: $\alpha_0$, which will induce a $3$-grading on $\ml$ similar to the one we had

532: for $\lie$:

533: \begin{equation}\label{grading2}

534: \ml=\ml_{-1}\oplus\ml_{0}\oplus\ml_{1}.

535: \end{equation}

536: 

537: Then the space $\ml_1$ will have a structure of a Jordan algebra.

538: 

539: One way

540: to prove it is by considering a compact real form of (\ref{grading2}) and

541: noticing that it produces a Hermitian symmetric space, which is of tube

542: type (see \cite[Section 4]{clerc} for details).

543: 

544: The structure of a Jordan algebra on $\ml_1$

545: can be described in purely algebraic terms,

546: see \cite{tits}, or \cite[Lemma~$4$]{kac}, which is a more accessible

547: reference.

548: 

549: 

550: \medskip

551: To summarize, given a cubic Jordan algebra $\J$, the Freudenthal

552: construction produces a group $\gr$ and its module $\m$ with

553: $\gr$-invariant quartic and symplectic forms. Also there exists a

554: simple Lie algebra $\lie\ (\ne A_n, C_n)$ with grading (\ref{grading}),

555: which yields the same data (and $\m=\lie_1)$. Moreover the subalgebra $\ml

556: \subset\lie_0$ possesses grading (\ref{grading2}) such that $\ml_1$

557: has the structure of a Jordan algebra isomorphic to $\J$.

558: 

559: 

560: \subsection{Organization of the paper}

561: We provide basic information about split

562: composition algebras and their integral structures

563: in the beginning of Section~$2$. Then we give basic information about Jordan

564: algebras.

565: We present the Springer construction of cubic Jordan algebras and

566: describe the three main examples of Jordan algebras possessing

567: an admissible cubic form (Subsection~$\ref{ss:ex}$). In this paper

568: we are mainly interested in cubic Jordan algebras $\h\calg$ of

569: $3\times 3$ Hermitian matrices over composition algebras

570: (Example~$\ref{ex:jcubic}$).

571: In Subsection~$\ref{ss:33herm}$ we describe the

572: groups $\str\J$ and $\npg\J$ associated with these algebras, and describe

573: orbits in $\h\calg$ under the action of the norm-preserving group

574: $\npg{\h\calg}$ (Proposition~$\ref{prop:np-o}$).

575: We conclude Section~$2$ with the description of the integral structures in

576: Jordan algebras of Subsection~$\ref{ss:ex}$, and the description of

577: integral orbits in $\h\calg$ (Theorem~$\ref{np-orbits}$).

578: \medskip

579: 

580: We begin Section~3 with the description of the Freudenthal construction.

581: Given a cubic Jordan algebra $\J$, this construction produces a group

582: $\gr$ and its representation $\mj$.

583: We identify groups $\gr$  and its representations $\mj$

584: in Proposition~$\ref{prop:gener}$ and Remark~$\ref{rem:semis}$, see also

585: Example~$\ref{ex:sl6}$.

586: 

587: We present the concept of rank of elements in the module $\mj$ in

588: Subsection~$\ref{ss:rank}$, and we use it in the classification of

589: $\gr$-orbits in $\m$ in Subsection~$\ref{ss:canf}$

590: (Theorem~$\ref{thm1}$). The key step in our approach is the computation of

591: Lemma~$\ref{lcomp}$, which allows to do a complete reduction of elements of

592: $\m$ using purely algebraic considerations.

593: 

594: Most of the assertions in Subsection~$\ref{ss:canf}$ are proved under the

595: assumption that the ground field $F$ is an arbitrary field of

596: characteristic $\ne 2,3$. The restriction char\,$F\ne 2$, is essential

597: in the paper since it is used in many intermediate computations (see also

598: Remark~$\ref{badred}$). As for characteristic $3$, it appears that this

599: restriction may be dropped without impairing the statements. However this

600: assumption was made in \cite{b}, which is our main reference concerning

601: the Freudenthal construction, and therefore we had to incorporate it in

602: our considerations.

603: 

604: \medskip

605: Section~$3$ may also be viewed as a testing ground (or a simpler version) of

606: the techniques that we apply to studying the integral case in Section~$4$.

607: We introduce the integral version of the Freudenthal construction  in

608: Subsection~$\ref{ss:intstr}$. We develop a reduction procedure for

609: elements of $\mz$, which is somewhat similar to the reduction of

610: $n\times n$ matrices with integer entries under the elementary row and

611: column transformations. This is done in Lemma~$\ref{bl-fz}$, which is one

612: of the main technical results in Section~4. This lemma does not provide a

613: complete reduction of the elements in $\mz$, but it is sufficient, for

614: example, to describe generators of the group $\grz$ (Proposition~$\ref{genz})$

615: in a way similar the case of a field.

616: 

617: Our elementary approach does not seem to be sufficient for the complete

618: description of orbits in $\mz$, and we restrict our attention to the

619: orbits of projective elements. This concept was introduced by M.Bhargava

620: \cite{bh-phd, bh1}

621: in the context of orbit spaces associated with ideal classes in quadratic

622: rings. In Subsection~$\ref{ss:proj}$ we show how this concept may be

623: transferred to elements of the modules $\mz$. We also provide a simple

624: test for the projectivity of elements in terms of the quartic invariant

625: of the module $\mz$ (Corollary~$\ref{cor:nabla}$,

626: Remark~$\ref{rem:nabla}$).

627: 

628: In Subsection~$\ref{ss:fred}$ we explain how one can do the complete

629: reduction for the projective elements (Lemma~$\ref{red2}$) and summarize

630: all the previous results in the proof of the main result of the paper

631: on the structure of orbits of the projective elements

632: (Theorem~$\ref{thm:main}$).

633: 

634: In the last subsection we analyze degenerate orbits of elements of $\mz$.

635: We are able to obtain a complete classification of orbits of elements of

636: rank $1$ and $2$. These results complement results of the previous

637: subsection; they are stated in Theorem~$\ref{thm2}$.

638: 

639: %

640: \medskip

641: We conclude the paper with an Appendix, where we describe a link between

642: the Freudenthal construction and Gauss's law of composition of quadratic

643: forms via the Cube Law of M.~Bhargava.

644: 

645: 

646: \subsection{Notation and conventions}

647: 

648: We work over a ground field $F$ of characteristic $\ne 2,3$.

649: The vector space of $n\times n$ matrices over $F$ is denoted by $M_n(F)$.

650: The identity matrix in $M_n(F)$ is denoted by $I_n$. For an arbitrary

651: vector space $V$ over $F$, the symbol $\id_V$ denotes the identity linear

652: transformation in $\End_F(V)$.

653: 

654: We use the word {\em space} to denote a set with an additional structure.

655: Depending on a context such a space may be a vector space, a

656: $\zet$-module, or a set of orbits under the action of a group

657: (an {\em orbit space}).

658: 

659: Many spaces that we consider have a certain integral structure introduced

660: in the paper. For a space~$V$,

661: the corresponding integral structure will be denoted by $V_\zet$.

662: Very often such spaces will arise as spaces of representation

663: of certain algebraic groups. In such cases $G(\zet)$ will denote the

664: set of $\zet$-rational points of $G$ with respect to the integral

665: structure in $V_\zet$. This is an {\em integral form\/} of the group $G$.

666: 

667: We will occasionally make references to simple split finite-dimensional

668: Lie algebras (algebraic groups) and their root systems.

669: The labeling of their simple roots, fundamental weights, etc. corresponds

670: to that of \cite{b1}.

671: 

672: 

673: \bigskip

674: {\bf Acknowledgements.}

675: The author is grateful to E.~Zelmanov and M.~Racine for their help and

676: encouragement. Many thanks to

677: B. Allison,

678: M.~Bhargava,

679: O.~Loos,

680: E.~Neher

681: for helpful discussions and comments on the content of the paper.

682: The author is also grateful to M.~Bhargava for providing a copy of his

683: manuscript~\cite{bh1}.

684: 

685: 

686: 

687: 

688: \section{Cubic Jordan algebras and associated structures}

689: \label{s:jordan}

690: 

691: The Freudenthal construction that we will describe in Section~\ref{sec:FC}

692: was originally introduced in the case of the $27$-dimensional exceptional

693: Jordan algebra. We are going to apply this construction to other examples

694: of the so called cubic Jordan algebras, and we introduce the appropriate

695: concepts in this section.

696: 

697: The basic information concerning Jordan algebras is given in

698: Subsection~$\ref{ss:jbasic}$. However, considering the Freudenthal

699: construction, we will not make much use of the Jordan product $\jprod$.

700: Instead, we will be looking at the cubic form $N$, the trace form and

701: the trace bilinear form, the ``sharp" operation $\#$ and its linearization

702: $\times$.

703: 

704: All of these operations (as well as the Jordan product $\jprod$) in a cubic

705: Jordan algebra may be defined starting with an admissible cubic form.

706: We will give the appropriate construction in Subsection~$\ref{ss:sc}$,

707: and provide explicit examples of admissible cubic forms (and Jordan

708: algebras) in Subsection~$\ref{ss:ex}$.

709: 

710: 

711: \subsection{Composition algebras and their integral structures}

712: \label{ss:compint}

713: 

714: A (not

715: necessarily associative) algebra $\calg$ with a unit element

716: is called a {\em composition} algebra

717: if it has a non-degenerate quadratic form $\ncomp$ ({\em the norm})

718: which permits composition, i.e.,

719: $$%

720: \ncomp(xy)=\ncomp(x)\ncomp(y),\qquad

721: \mbox{for all }x,y\in\calg.

722: $$%

723: 

724: A famous theorem of Hurwitz states that

725: such an algebra is always finite-dimensional,  %

726: and moreover,

727: its dimension can only be equal to $1, 2, 4$, or $8$

728: (see, e.g.,  \cite[Section~II.2.6]{atoj} or \cite[Theorem~1.6.2]{SpV}).

729: In this paper we

730: restrict our attention to {\em split} composition algebras only (a

731: composition algebra is split, if it contains non-zero elements of zero

732: norm; in that case the quadratic form $\ncomp$ has maximal Witt index).

733: For a field $F$ there is a unique up to isomorphism split composition

734: algebra of a given dimension $(2,4,$ or $8)$ \cite[Theorem 1.8.1]{SpV}.

735: We will call it the algebra of split {\em binarions, quaternions, octonions},

736: and denote it by

737: $\bin, \quat, \oct$, respectively.

738: We will use symbol $\calg$ to denote an arbitrary fixed composition

739: algebra.

740: 

741: Next we will present the construction of the three split

742: composition algebras and show how to define an integral structure in each

743: of them.

744: 

745: Our starting point for constructing composition algebras will be the

746: algebra of $2\times 2$ matrices over the field $F$. This algebra has

747: dimension $4$ and in fact it can be taken as a model for the algebra of

748: split quaternions $\quat$. The quadratic form $\ncomp$ in this algebra

749: is the usual determinant of $2\times 2$ matrices $\det$. The composition

750: property for $\ncomp$ is just the multiplicativity of the determinant.

751: 

752: One can define the concepts of {\em trace} and {\em conjugation} of

753: elements in a composition algebra. In the case of quaternions they are

754: the trace and the symplectic involution of a $2\times 2$ matrix:

755: \begin{equation}

756: \tcomp (\bfa)=a+d,\qquad

757: \conj{\bfa}=\maketwo{d}{-b}{-c}a,\qquad

758: {\rm for}\

759: {\bfa }=\maketwo{a}bcd\in \quat.

760: \end{equation}

761: 

762: One can define the algebra of binarions $\bin$ to be the subalgebra of

763: $\quat$, which consists of the diagonal elements

764: \begin{equation}\label{binar}

765: \bin=\left\{\left.

766: \maketwo{a}00d\right|

767: a,d\in F\right\}

768: \end{equation}

769: 

770: The quadratic form, trace, and the conjugation operation in $\bin$ are

771: induced by those in $\quat$.

772: 

773: We may consider the ground field $F$ to be embedded in the binarions

774: \begin{equation}

775: F\cong \left\{\left.\maketwo{a}00a\right|a\in F\right\}

776: \subset\bin,

777: \end{equation}

778: and consider the induced structure of a one-dimensional composition algebra.

779: %

780: 

781: \medskip

782: Finally, we define the algebra of octonions using the Cayley-Dickson

783: duplication process \cite{atoj, SpV}. $\oct$ is defined to be the

784: direct sum $\quat\oplus\quat$, and we write an arbitrary octonion with the

785: aid of a formal variable $v$ (the ``imaginary unit") as

786: \begin{equation}

787: {\bfa}+\bfb v,\qquad \bfa , \bfb\in \quat.

788: \end{equation}

789: 

790: One then defines

791: $$

792: ({\bf a+b}v)\cdot ({\bf c+d}v)\defeq

793: {\bf (ac-\conj{d}b)+(da+b\conj{c})}v,

794: \qquad {\bf a,b,c,d}\in \quat.

795: $$

796: \begin{equation}

797: \ncomp({\bf a+b}v) \defeq \det({\bf a})+\det({\bf b}),\quad

798: \tcomp({\bf a+b}\it v) \defeq \tcomp({\bf a}), \quad

799: \conj{{\bf a+b}v} \  \defeq \conj{{\bf a}}-{\bf b}v.

800: \end{equation}

801: 

802: These operations turn $\oct$ into a composition algebra.

803: 

804: We note that the multiplication of binarions is both associative and

805: commutative; the multiplication of quaternions is associative, but not

806: commutative; and, finally, we lose both the associative and commutative

807: property of the multiplication of the octonions.

808: 

809: We identify elements of the ground field $F$ with the appropriate

810: multiples of the identity element in the composition algebra. These are

811: the only elements which are fixed under the conjugation.

812: 

813: 

814: \bigskip

815: 

816: To define an integral structure in these algebras, we again start with

817: the quaternions. We define integral quaternions to be the $2\times 2$

818: matrices with integral entries

819: \begin{equation}

820: \quatz=

821: \left\{\left.

822: \maketwo{a}bcd\right|

823: a,b,c,d\in \zet\right\}.

824: \end{equation}

825: 

826: This integral structure is extended to the binarions and octonions in a

827: natural way

828: \begin{equation}

829: \binz=

830: \left\{\left.

831: \maketwo{a}00d\right|

832: a,d\in \zet\right\},\qquad

833: \octz=

834: \Bigl\{{\bfa}+\bfb v\Bigr|\ \bfa , \bfb\in \quatz

835: \Bigr\}.

836: \end{equation}

837: 

838: It is easy to see

839: that each of these $\zet$-modules has the structure of a composition

840: algebra (over~$\zet$), and the trace and the norm defined on them have

841: integer values.

842: 

843: 

844: \subsection{Basics of Jordan algebras}

845: \label{ss:jbasic}

846: 

847: We are going to present basic definitions and constructions concerning

848: Jordan algebras in this subsection. The classical reference for the

849: subject is N.~Jacobson's book \cite{ja}. A modern exposition of

850: the theory of Jordan algebras may be found in the recent monograph by

851: K.~McCrimmon \cite{atoj}.

852: 

853: \begin{defn}

854: {\rm

855: A (linear) {\em Jordan algebra} $\J$ over a field $F$ of char$\ne 2$ is a vector

856: space over $F$ with a bilinear product $\jprod$

857: (the {\em Jordan product}) satisfying the following two axioms:

858: \begin{equation} \label{j-ax}

859: x\jprod y =y\jprod x;\qquad

860: x^2\jprod (x\jprod y) = x\jprod (x^2 \jprod y)\qquad

861: {\rm \ for \ }x,y\in \J

862: \end{equation}

863: ($x^2$ is defined as $x\jprod x$).

864: }

865: \end{defn}

866: The Jordan product is commutative by definition, but it does not have to satisfy

867: the associative law.

868: 

869: 

870: A prototypical example of a Jordan algebra is the algebra $A^+$

871: obtained from any associative algebra $A$ (e.g., a matrix algebra)

872: by defining the Jordan product to be

873: \begin{equation}  \label{j-assoc}

874: x\jprod y =\frac{1}{2}(xy+yx).

875: \end{equation}

876: Here and below the product $xy$ represents the multiplication in the

877: original (associative) algebra~$A$.

878: 

879: \medskip

880: We will be mostly interested in the so-called cubic Jordan

881: algebras, i.e., Jordan algebras,

882: in which every element satisfies a cubic polynomial equation.

883: This concept can be introduced via the concept of {\em generic norm} and

884: {\em generic minimal polynomial} \cite[Section VI.3]{ja}.

885: 

886: However in this paper we will use a different approach. Namely, we will

887: consider certain examples of Jordan algebras,

888: which are constructed from a vector

889: space with a cubic form. The details of the process are given in the

890: following subsection.

891: 

892: \subsection{The Springer construction of cubic Jordan algebras}

893: \label{ss:sc}

894: 

895: In this subsection we will present a construction of a class of Jordan

896: algebras obtained from a cubic form on a vector space, known as the

897: Springer construction. Our exposition will follow

898: \cite[Section~I.3.8]{atoj}. The original references are

899: \cite{spr}, \cite[Section 4]{mc-sp}. We are still working under the

900: assumption that char$F\ne 2,3$, and we will use it occasionally to

901: simplify the constructions.

902: 

903: \bigskip

904: A cubic form $\mnorm$ on a vector space $V$ over $F$ (char\,$F\ne 2,3$) is a map

905: $\mnorm:V\to F$ such that

906: \begin{itemize}

907: \item $\mnorm(\alpha x)=\alpha^3 \mnorm(x)$ for $\alpha\in F, x\in V$;

908: \item $\mnorm(x,y,z)$ is a trilinear function $V\times V \times V\to F$;

909: \end{itemize}

910: where $\mnorm(x,y,z)$ is the full linearization of $\mnorm$ given by

911: $$

912: \mnorm(x,y,z)=\frac{1}{6}

913: \Bigl(\mnorm(x+y+z)-\mnorm(x+y)-\mnorm(x+z)-\mnorm(y+z)+\mnorm(x)+\mnorm(y)+\mnorm(z)\Bigr).

914: $$

915: In particular, we have

916: $$

917: \mnorm (x,x,x) = \mnorm (x).

918: $$

919: 

920: We say that $c\in V$ is a {\em basepoint} for $\mnorm$, if $\mnorm(c)=1$. One then

921: can define the following maps

922: \begin{itemize}

923: \item

924: a linear map (trace) $V\to F$: $\tr(x)=3\,\mnorm(c,c,x)$;

925: \item

926: a quadratic map $V\to F$: $S(x)=3\mnorm(x,x,c)$;

927: \item

928: a bilinear map $V\times V\to F$: $S(x,y)=6\mnorm(x,y,c)$;

929: \item

930: a trace bilinear  form $V\times V\to F$: $(x,y)=\tr(x)\tr(y)-S(x,y)$.

931: \end{itemize}

932: 

933: In particular, we have

934: $$

935: \mnorm(c)=1;\qquad

936: S(c)=3;\qquad

937: \tr(c)=3.

938: $$

939: 

940: \begin{defn}

941: {\em

942: A cubic form with a basepoint $(\mnorm,c)$ on a finite-dimensional vector space

943: $V$ over a field $F$ of char$\ne 2,3$ is said to be {\em admissible}

944: or a {\em Jordan cubic}, if

945: \begin{enumerate}

946: \item[(1)]

947: $\mnorm$ is nondegenerate at the basepoint $c$ in the sense that

948: the trace bilinear form $(x,y)$ is nondegenerate;

949: \item[(2)]

950: The quadratic {\em adjoint} (or {\em sharp}) map $V\to V$,

951: defined uniquely by $(x^\#, y)=3\mnorm(x,x,y)$, satisfies the adjoint identity:

952: \begin{equation}\label{adid}

953: (x^\#)^\#=\mnorm(x)x.

954: \end{equation}

955: %

956: %

957: \end{enumerate}

958: We will also use the term {\em Jordan cubic} when referring to the

959: associated Jordan algebra (see Proposition~\ref{prop:jcubic}).

960: }

961: \end{defn}

962: 

963: The following relation holds for all $x$ in $V$:

964: \begin{equation}\label{trace-spur}

965: \tr(x^\#)=S(x).

966: \end{equation}

967: 

968: \bigskip

969: We define the linearization of the sharp map

970: \begin{equation}

971: x\cross y = (x+y)^\#-x^\#-y^\#.

972: \end{equation}

973: Note that

974: $$

975: x\cross x = 2 x^\#.

976: $$

977: 

978: 

979: 

980: \begin{prop}\label{prop:jcubic}

981: {\em \cite[Section I.3.8]{atoj}}

982: Every vector space with an admissible cubic form gives rise to a Jordan

983: algebra with unit $\one=c$ and the Jordan product given by

984: \begin{equation}\label{j-prod}

985: x\jprod y =

986: \frac{1}{2}\Bigl( x\cross y+\tr(x)y+\tr(y)x-S(x,y)\one \Bigr).

987: \end{equation}

988: Every element of this Jordan algebra satisfies the cubic polynomial:

989: \begin{equation} \label{3poly}

990: x^3-\tr(x)x^2+S(x)x-\mnorm(x)\one=0.

991: \end{equation}

992: We also have

993: $$

994: x^\#=x^2-\tr(x)x+S(x)\one.

995: $$

996: \end{prop}

997: 

998: Taking the trace of the last expression and then using~$(\ref{trace-spur})$,

999: one gets

1000: $$

1001: \tr(x^2)=\tr(x)^2-2S(x),

1002: $$

1003: 

1004: which linearizes to

1005: \begin{equation} \label{traces}

1006: \tr(x\jprod y)=(x,y).

1007: \end{equation}

1008: 

1009: \medskip

1010: The following simple example illustrates the concepts introduced above.

1011: 

1012: \begin{example}\label{ex:mat3}

1013: {\rm

1014: Let $V$ be the vector space $M_{3}(F)$ of $3\times 3$ matrices over $F$.

1015: We let the cubic form $\mnorm$ be the determinant, and we let $c$ be the

1016: identity matrix.

1017: 

1018: Then the linear map $\tr$ is the regular trace of matrices. The sharp map

1019: $x^\#$ produces the classical adjoint (transposed cofactor) matrix of $x$.

1020: Equation (\ref{j-prod}) yields the Jordan product, which coincides with the

1021: product in the Jordan algebra $M_{3}(F)^+$ given by (\ref{j-assoc}):

1022: $$

1023: x\jprod y=\frac{1}{2}(xy+yx).

1024: $$

1025: The trace bilinear form $(x,y)$ is equal to $\tr(x\jprod y)$.

1026: It coincides in this

1027: example with the standard trace form in the matrix algebra.

1028: Finally, we notice that the equation (\ref{3poly}) becomes just the~

1029: Cayley-Hamilton equation for $3\times 3$ matrices.

1030: }

1031: \ee

1032: \end{example}

1033: 

1034: 

1035: \subsection{Three main examples}

1036: \label{ss:ex}

1037: 

1038: We will provide several more examples of cubic Jordan algebras in this

1039: subsection.

1040: 

1041: 

1042: The following example is a simplified version of the {\em reduced cubic

1043: factor example} \cite[Section~I.3.9]{atoj}.

1044: \begin{example}\label{ex:jcubic}

1045: {\rm

1046: Let $\calg$ be a composition algebra over a field $F$

1047: with a quadratic form $\ncomp$ and an involution $\bar{\ }$.

1048: 

1049: We let $V$ be the space $\h{\calg}$ of $3\times 3$ Hermitian matrices over $\calg$.

1050: An arbitrary element $A$ in $\h{\calg}$ has the form

1051: $$

1052: A=\makeherm{a}bc{\bf x}{\bf y}{\bf z},\qquad

1053: \begin{array}{l}

1054: \mbox{where }\ a,b,c \in \F, \\

1055: \mbox{and }{\bf x,y,z}\in\calg.

1056: \end{array}

1057: $$

1058: The basepoint $c$ is defined to be the identity (diagonal) matrix in

1059: $\h\calg$, and the cubic form

1060: $$

1061: N:\h{\calg}\to F

1062: $$

1063: is given by the expression

1064: reminiscent of the regular determinant

1065: \begin{equation} \label{cubic1}

1066: \mnorm(A) = \ abc-a\, {\bf x\conj{x}}-b\, {\bf y\conj{y}}-c\, {\bf z\conj{z}}+

1067: {\bf (xy)z+\bar{\bf z}{(\bar{\bf y}\bar{\bf x})}}.

1068: \end{equation}

1069: 

1070: The trace form $\tr$, as defined in the previous section from $\mnorm$,

1071: coincides with the regular trace $\tr(A)=a+b+c$ and

1072: $\btr(A,B)=\tr\Bigl(\frac{1}{2}(AB+BA)\Bigr)$.

1073: The ``sharp" operation produces the regular adjoint matrix:

1074: $$

1075: A^\#=

1076: \left(

1077: \begin{array}{lll}

1078: 

1079:    bc-\ncomp({\bf x}) \quad                       & \conj{\bf y}\ \conj{\bf x}-c\,{\bf z} \quad   & {\bf z\, x}-b\,\conj{{\bf y}} \\

1080:    {\bf x}\ {\bf y}-c\,\conj{\bf z}     & ac-\ncomp({\bf y})                          & {\bf \conj{z}\ \conj{y}} -a\,{\bf x}\\

1081:    {\bf \conj{x}\, \conj{z}}-b\,{\bf y} & {\bf y \, z} -a\,{\bf x}                &  ab-\ncomp({\bf z})

1082: \end{array}

1083: \right).

1084: $$

1085: 

1086: The cubic form $\mnorm$ defined by $(\ref{cubic1})$ is admissible.

1087: }\ee

1088: \end{example}

1089: 

1090: 

1091: We note that Example \ref{ex:mat3} can be viewed as

1092: a special case of Example \ref{ex:jcubic}

1093: as explained in the following

1094: 

1095: \begin{rem}\label{rem:hbm3}

1096: {\rm

1097: If the composition algebra $\calg$ is the algebra of split binarions

1098: $\bin$, then an arbitrary element of $\h\bin$ has the form

1099: (see~$(\ref{binar})$)

1100: $$

1101: \makethree

1102: {a}                 {\bfa_{12}}         {\bfa_{13}}

1103: {\conj{\bfa_{12}}}  {b}                 {\bfa_{23}}

1104: {\conj{\bfa_{13}}}  {\conj{\bfa_{23}}}  {c}

1105: ,\qquad

1106: \mbox{where }

1107: \bfa_{ij}=\maketwo{a_{ij}}{\scriptstyle 0}{\scriptstyle 0}{a_{ji}}\in

1108: \bin,\quad

1109: a,b,c, \in F.

1110: $$

1111: 

1112: We then define a linear map $\h\bin\to M_3(F)$

1113: \begin{equation}\label{bin3iso}

1114: \makethree

1115: {a}                 {\bfa_{12}}         {\bfa_{13}}

1116: {\conj{\bfa_{12}}}  {b}                 {\bfa_{23}}

1117: {\conj{\bfa_{13}}}  {\conj{\bfa_{23}}}  {c}

1118: %

1119: \longmapsto

1120: \makethree

1121: {a}       {a_{12}}  {a_{13}}

1122: {a_{21}}  {b}       {a_{23}}

1123: {a_{31}}  {a_{32}}  {c}.

1124: \end{equation}

1125: 

1126: It is easy to see that this isomorphism of vector spaces

1127: is a norm isometry, where the cubic form in $M_3(F)$ is

1128: the regular determinant.

1129: Hence $(\ref{bin3iso})$ defines an isomorphism of

1130: Jordan algebras.

1131: \ee

1132: }

1133: \end{rem}

1134: 

1135: \begin{example}\label{ex:quad}

1136: {\rm

1137: \cite[Section~$4$]{mc-sp}

1138: Let $Q_n$ be a vector space of dimension $n$ over a field $F$,

1139: and let $B_0$ be a

1140: non-degenerate quadratic form on $Q_n$ with a basepoint $c_0$

1141: (i.e., $B_0(c_0)=1$). We form a vector space $V$ by

1142: taking the direct sum of a copy of the

1143: ground field $F$ and $Q_n$:

1144: $$

1145: V=F \oplus Q_n.

1146: $$

1147: 

1148: We then define a cubic form $\mnorm$ on $V$ by

1149: $$

1150: \mnorm(\alpha, x_0)=\alpha\, B_0(x_0), \qquad

1151: \mbox{for }\alpha \in F, x_0\in Q_n.

1152: $$

1153: We let $c=(1,c_0)$. A simple verification

1154: shows that $c$ is a basepoint for $\mnorm$

1155: and that the cubic form $\mnorm$ is admissible.

1156: 

1157: In particular, the following formulas hold for $x,y \in V$:

1158: \begin{eqnarray*}

1159: x^\# &=& \Bigl(B_0(x_0),\ \alpha x_0^*\Bigr),\\

1160: (x,y)&=& \alpha\beta + B_0(x_0^*, y_0),

1161: \end{eqnarray*}

1162: where $x=(\alpha, x_0), y=(\beta, y_0)$, $x_0^*=B_0(x_0, c_0)c_0-x_0$, and

1163: $B_0(\cdot,\cdot)$ is a linearization of the quadratic form $B_0(\cdot)$:

1164: $$

1165: B_0(u,v)=B_0(u+v)-B_0(u)-B_0(v).

1166: $$

1167: }\ee

1168: \end{example}

1169: 

1170: \begin{example}\label{ex:f1}

1171: {\rm

1172: We let $V$ be the one-dimensional vector space $V=F$. We define a cubic

1173: form $\mnorm$ on $V$ by

1174: $$

1175: N(\alpha)=\alpha^3, \qquad

1176: \alpha\in V

1177: $$

1178: with the obvious choice of the base point $c=1\in F$. We have

1179: $\tr(\alpha)=3\alpha$, $(\alpha, \beta)=3\alpha\beta$,

1180: $\alpha^\#=\alpha^2$ for $\alpha, \beta\in F$, and evidently

1181: $N$ is an admissible cubic form.

1182: }\ee

1183: \end{example}

1184: 

1185: 

1186: 

1187: \subsection{Two groups associated to cubic Jordan algebras}

1188: \label{ss:33herm}

1189: 

1190: Next we will introduce two groups associated to a Jordan algebra $\J$. The

1191: definitions, due to N.~Jacobson, are valid for all finite-dimensional

1192: Jordan (or power-associative) algebras, but we will use them only for

1193: Jordan algebras with a cubic form.

1194: 

1195: \begin{defn}\label{def-groups}

1196: 

1197: The {\em norm-preserving group}

1198: $$

1199: \npg\J=\Bigl\{g\in GL(\J)\ |\  \mnorm(g A)=\mnorm(A)\mbox{ for all }A\in \J\Bigr\}

1200: $$

1201: is the group of all invertible $F$-linear transformations of the

1202: vector space $\J$, which preserve the norm~$\mnorm$.

1203: 

1204: Similarly we define the {\em group of norm similarities} or

1205: the {\em structure group}

1206: $$

1207: \str\J=\Bigl\{g\in GL(\J)\ |\  \mnorm(g A)=\chi(g)\, \mnorm(A)\mbox{ for all }A\in \J\Bigr\},

1208: $$

1209: where $\chi(g)$ is a scalar in $F$, which depends on the group element $g$

1210: only.

1211: \end{defn}

1212: 

1213: We have the obvious inclusion $\npg\J \subset \str\J$.

1214: 

1215: \bigskip

1216: 

1217: 

1218: \begin{rem}\label{groups}

1219: {\rm

1220: Here we will provide a description of these groups for cubic Jordan

1221: algebras of the form $\h\calg$

1222: (\cite[Ch.~$14$]{sp-alg}, see also \cite{J3}, \cite[VI.7--9]{ja}).

1223: %

1224: %

1225: 

1226: \begin{itemize}

1227: \item[(i)]

1228: {\bf Case} $\J=\h{F}$.

1229: 

1230: $\str\J$ is the group of transformations of the form

1231: $$

1232: X\mto \gamma AXA^t,

1233: $$

1234: 

1235: where $X\in\h{F}$, $A\in GL_3(F)$, $\gamma\in F^\mult$, $A^t$ is the transpose of $A$.

1236: %

1237: 

1238: $\npg\J$ consists of transformations for which

1239: $\gamma^3(\det A)^2=1$.

1240: 

1241: 

1242: \item[(ii)]

1243: {\bf Case} $\J=\h{\bin}$.

1244: 

1245: In this case $\J$ is isomorphic to the Jordan algebra of all $3\times 3$

1246: matrices over $F$ (Remark~$\ref{rem:hbm3}$).

1247: 

1248: The group $\str\J$ is generated by the transformations of the form

1249: $$

1250: \eta(A,B): X\mto AXB^{-1}\quad

1251: \mbox{ and }\quad

1252: t: X\mto X^t,

1253: $$

1254: 

1255: where $X\in M_3(F)$, $A,B\in GL_3(F)$, and $X^t$ denotes the transpose of $X$.

1256: 

1257: $\eta(A,B)$ acts trivially iff $A=B=\alpha\id, \ \alpha\in F,\alpha\ne 0$.

1258: 

1259: $\npg\J$ consists of transformations $\eta(A,B)$ for which $\det A=\det B$.

1260: 

1261: %

1262: %

1263: 

1264: We will let $\str \J\cc$ denote the subgroup of $\str\J$ that

1265: consists of the transformation $\eta(A,B)$. It is a subgroup of index two

1266: in $\str\J$, and we have $\str\J\cong\str\J\cc\rtimes \{1, t\}$.

1267: Similarly we define $\npg\J\cc=\npg\J \cap \str\J\cc$.

1268: 

1269: \item[(iii)]

1270: {\bf Case} $\J=\h{\quat}$.

1271: 

1272: We have the isomorphisms of Jordan algebras

1273: $\h{\quat}\cong {\cal H}(M_{6}(F), {\rm symp})$ (the $6\times 6$ symplectic

1274: symmetric matrices) (see, e.g., \cite[p. 65]{J3}).

1275: 

1276: The group $\str\J$ is the group of transformations of the form

1277: $$

1278: X\mto \gamma AXA^{\rm symp}

1279: $$

1280: 

1281: where $X\in{\cal H}(M_{6}(F), {\rm symp})$,

1282: $A\in GL_6(F)$, $\gamma\in F$, $\gamma\ne 0$, $A^{\rm symp}$ is the transpose of $A$

1283: with respect to the symplectic involution.

1284: 

1285: $\npg\J$ consists of transformations for which $\gamma^3 (\det A)^2=1$.

1286: 

1287: \item[(iv)]

1288: {\bf Case} $\J=\h{\oct}$.

1289: 

1290: In this case we have that $\npg{\J}$ is a simply connected

1291: simple algebraic group of type $E_6$, whose center is

1292: isomorphic to the group $\bfmu _3(F)$

1293: of third roots of unity in $F$.

1294: 

1295: The group $\str\J$ in this case is equal to the product

1296: $\npg\J \cdot (F^\mult\id_\J)$ with $\npg\J \cap F^\mult\id_\J=\bfmu_3(F)\id$.

1297: \end{itemize}

1298: 

1299: Later we will view groups $\str\J$ and $\npg\J$ as the group of

1300: $F$-rational points of the algebraic groups $\astr \J$ and $\anpg\J$,

1301: respectively. These group are connected in all cases, except $\J=\h\bin$.

1302: For $\J=\h\bin$, the group $\astr{\J}$ (respectively, $\anpg{\J}$)

1303: has two connected components; and the set of $F$-rational points of the

1304: component of the identity coincides with $\str{\J}\cc$ (respectively,

1305: $\npg{\J}\cc$).

1306: \ee

1307: }

1308: \end{rem}

1309: 

1310: The following definition generalizing the concept of rank for the usual

1311: $3\times 3$ matrices goes back to N.~Jacobson.

1312: 

1313: \begin{defn}\label{def:jrank}

1314: Let $\J$ be a cubic Jordan algebra and let $\mnorm$ be the cubic form on $\J$.

1315: The {\em rank} of an arbitrary element $A\in\J$ is an integer between zero and

1316: three, which is defined by the following relations:

1317: \begin{itemize}

1318: \item

1319: \rank$A =3$ iff \ $N(A)\ne 0$;

1320: \item

1321: \rank$A \le 2$ iff \ $N(A)=0$;

1322: \item

1323: \rank$A \le 1$ iff \ $A^\#=0$;

1324: \item

1325: \rank$A=0$ iff \ $A=0$.

1326: \end{itemize}

1327: \end{defn}

1328: 

1329: It is known that the rank is invariant under the action of the

1330: groups $\npg\J$ and $\str\J$, see, e.g., \cite[Section~$2$]{J3}.

1331: 

1332: \medskip

1333: We will conclude this subsection by describing orbits in $\h\calg$

1334: under the action of the \np\ group $\npg{\h\calg}$.

1335: The classification is based on the concept of rank

1336: and is analogous to the classification of orbits in $M_3(F)$

1337: under the action of the elementary row and column transformations.

1338: 

1339: \begin{prop}\label{prop:np-o}

1340: 

1341: \

1342: \begin{enumerate}[\bf (i)]

1343: \item

1344: Let $\calg$ be the split composition algebra over a field $F$,

1345: $\calg=\bin, \quat, \oct$, char $F\ne 2,3$.

1346: Let $\h{\calg}$ be the cubic Jordan algebra

1347: of $3\times 3$ Hermitian matrices over $\calg$.

1348: 

1349: Then the group $\npg{\h{\calg}}$ acts transitively on the

1350: sets of elements of rank $1$ and $2$. In the case of rank\/ $3$,

1351: the group $\npg{\h{\calg}}$ acts transitively on the elements

1352: of a given norm $k$, $k\in F$, $k\ne 0$.

1353: 

1354: All these orbits are distinct and the union of these orbits and $\{0\}$ is

1355: $\h{\calg}$.

1356: \item

1357: If in addition every element of $F$ is a square,

1358: the same result holds for $\h{F}$.

1359: \end{enumerate}

1360: \end{prop}

1361: 

1362: It follows that the orbit representatives for the action of

1363: $\npg{\h{\calg}}$ may be chosen to be the following

1364: diagonal matrices:

1365: $$

1366: 0, \quad

1367: \makethree 1.. .0. ..0,\quad

1368: \makethree 1.. .1. ..0,\quad

1369: \makethree 1.. .1. ..k,\quad

1370: k\ne 0.

1371: $$

1372: 

1373: {\sc Proof.}

1374: 

1375: {\bf (i)}

1376: The statement of the theorem is obvious for matrices in

1377: $\h\bin\cong M_3(F)$ (Remark~$\ref{rem:hbm3}$).

1378: For other split composition algebras the statement was

1379: essentially known to N.~Jacobson, see e.g. \cite[Section~$2$]{J3}.

1380: Alternatively, one may view this proposition as a corollary (of the proof)

1381: of the main theorem of \cite{k}, see also Theorem~$\ref{np-orbits}$ below.

1382: 

1383: {\bf (ii)}

1384: The proof of (i) does not work in the case of Hermitian matrices over the

1385: ground field $F$. The result here depends on the arithmetic properties

1386: of the ground field. When $F$ satisfies the assumptions of (ii), the

1387: assertion is well known, see, e.g. \cite[Theorems~$6.5, 6.6$]{jba}.

1388: \ep

1389: 

1390: \subsection{Operations and identities in cubic Jordan algebras}

1391: \label{ss:id}

1392: 

1393: Given an arbitrary Jordan algebra $\J$, one can define the

1394: {\em Jordan triple product}

1395: \begin{equation}\label{jtp}

1396: \{x,y,z\}  =  (x\jprod y)\jprod z+ x\jprod (y\jprod z)-(x\jprod z)\jprod

1397: y,\qquad

1398: x,y,z\in\J.

1399: \end{equation}

1400: 

1401: When $\J\subseteq A^+$ for some associative algebra $A$, the Jordan triple

1402: product has a simple expression in terms of the associative operation in

1403: $A$:

1404: \begin{equation}\label{a:triple}

1405: \{x,y,z\}=\frac{1}{2}(xyz+zyx),\qquad

1406: x,y,z\in \J\subseteq A^+.

1407: \end{equation}

1408: This operation is very important for generalizations of Jordan algebras.

1409: We mentioned it here, since we need to introduce another operation

1410: $V_{x,y}:\J\to\J$ defined by

1411: \begin{eqnarray}\label{a:v}

1412: V_{x,y} (z) & = & \{x, y, z\},\qquad

1413: x,y,z\in \J.

1414: \end{eqnarray}

1415: We will use the operation $V_{x,y}$ in

1416: Subsections~$\ref{ss:rank}$ and~$\ref{ss:if}$ in the definition

1417: of rank of elements of the module~$\mj$.

1418: 

1419: \medskip

1420: 

1421: Next we list several identities which relate the triple product, the cross

1422: product, and the trace bilinear form in an arbitrary {\em cubic} Jordan

1423: algebra~$\J$.

1424: 

1425: It follows from $(\ref{3poly})$ that

1426: $$

1427: X\jprod X^\#=\mnorm(X)\,\one

1428: $$

1429: 

1430: for any element $X$ in $\J$.

1431: 

1432: 

1433: %

1434: 

1435: One can linearize this identity

1436: (see, e.g., \cite[Section~${2}$]{J3}, \cite[Section~5.2]{SpV})

1437: and get

1438: 

1439: \begin{eqnarray} \label{a:id1}

1440: \{X,Y,X\}+2Y\cross (X^\#)&=&(X,Y)X\\ \label{a:id2}

1441: (X,Z)Y+(Y,Z)X&=&2\{X,Z,Y\}+(X\cross Y)\cross Z\\ \label{a:id3}

1442: N(Y)X+(X,Y^\#)Y&=&(X\cross Y)\cross Y^\#\\ \label{a:id4}

1443: (X,Z^\#)X+\{Z,X^\#,Z\}&=&(X\cross Z)^\#.

1444: \end{eqnarray}

1445: 

1446: 

1447: One can define the * operation

1448: for an arbitrary element $s$ of the \np\ group

1449: $\npg\J$ by the relation:

1450: \begin{equation} \label{a:adj}

1451: \Bigl(\ s   (X),\ Y\ \Bigr) \ =

1452: \ \Bigl(\ X,\ s^* (Y)\ \Bigr) \qquad \mbox{for any }X,Y\in \J.

1453: \end{equation}

1454: 

1455: It satisfies the following identities:

1456: \begin{equation} \label{a:crossid}

1457: {s^{*-1}}(X\cross Y)= s (X) \cross s (Y) \qquad \mbox{for any }s\in\npg\J,\ X,Y\in\J.

1458: \end{equation}

1459: \begin{equation} \label{a:tripleid}

1460: s \Bigl(\{X,Y,Z\}\Bigr)= \Bigl\{\ s (X),\ {s^{*-1}}(Y),\ s (Z)\ \Bigr\} \qquad

1461: \mbox{for any }s\in\npg\J,\ X,Y,Z\in\J.

1462: \end{equation}

1463: 

1464: 

1465: \subsection{The integral case and orbits under the \np\ group}

1466: \label{ss:33int}

1467: 

1468: The integral structure $C_\zet$

1469: of the composition algebras $C$ in Subsection~$\ref{ss:compint}$ induces an

1470: integral structure in the spaces $\h{C_\zet}$.

1471: 

1472: We note that $\h{C_\zet}$ is not closed under the Jordan product

1473: because of the factor $\frac{1}{2}$, but

1474: it is easy to see from the expressions given in Example~$\ref{ex:jcubic}$,

1475: that the cubic form $\mnorm$, the trace $\tr$ and the trace bilinear form

1476: $\btr(\cdot, \cdot)$ take values in (and onto) $\zet$. We also have that

1477: $\h{C_\zet}$ is closed under the sharp operation $\#$, and

1478: $X\cross Y\in\h{C_\zet}$ for any $X,Y\in \h{C_\zet}$.

1479: 

1480: 

1481: We consider the \np\ groups defined in Subsection~$\ref{ss:33herm}$,

1482: and we look at

1483: their subgroups of elements which preserve the integral submodule

1484: $\h{C_\zet}$. This subgroup is an integral form of the appropriate group.

1485: We consider the action of each of these groups on the space of

1486: integral Hermitian matrices.

1487: 

1488: The structure of the orbits in $\h{C_\zet}$ under the action

1489: of the \np\ groups is described in

1490: Theorem~\ref{np-orbits} below. Before stating the theorem

1491: we will recall the following well-known

1492: 

1493: \begin{defn}

1494: {\rm

1495: We say that an $n\times n$ matrix $A$ is in the {\em Smith normal form},

1496: if $A$ is a diagonal matrix

1497: $$

1498: A=\diag{d_1, d_2}{\dots}{d_n},

1499: $$

1500: where

1501: $$

1502: \mbox{all }d_i\mbox{'s are integers, }\quad

1503: d_i \,|\, d_{i+1},\quad d_i\ge 0\quad

1504: \mbox{ for }1\le i \le n-1

1505: $$

1506: and all zeros on the diagonal are located in the lower right corner.

1507: }

1508: \end{defn}

1509: 

1510: \begin{theorem} \label{np-orbits}

1511: Let $\h{\calg_\zet}$ be the $\zet$-module of $3\times 3$ Hermitian matrices

1512: over the split composition {\em ring} $\calg_\zet$, $\calg_\zet=\bin_\zet,

1513: \quat_\zet, \oct_\zet$.

1514: 

1515: Then every element of $\h{\calg_\zet}$ is equivalent to an element in the

1516: {\em Smith normal form} under the action of the group $\npg{\h{\calg_\zet}}$.

1517: Distinct elements in the Smith normal form lie in distinct

1518: $\npg{\h{\calg_\zet}}$-orbits.

1519: \end{theorem}

1520: 

1521: {\sc Proof.}

1522: The assertion of this theorem when $C_\zet=\oct_\zet$

1523: was proved in \cite{k} (we used the term {\em canonical diagonal form}

1524: to represent Smith normal form).

1525: 

1526: The reasoning of that paper also applies in the case of integer

1527: quaternions and was stated as Corollary there \cite[p.~294]{k}.

1528: This assertion may also be stated in terms of integer skew-symmetric

1529: matrices (cf. Remark~\ref{groups}(iii)), see \cite[Theorem~IV.1]{newman}.

1530: 

1531: In case of integer binarions, one could also repeat the argument of

1532: \cite{k} to arrive to the same conclusion. An alternative way

1533: to prove the theorem is to notice that the isomorphism of

1534: Remark~$\ref{rem:hbm3}$ yields a $\zet$-linear norm isometry of

1535: $\zet$-modules

1536: $\h{\binz}$ and $M_{3}(\zet)$

1537: A consequence of this fact is that the action of the \np\

1538: group in $\h\binz$ can be expressed in terms of the elementary row and columns

1539: transformations of the regular $3\times 3$ matrices over $\zet$.

1540: Hence in the case of

1541: integral binarions the assertion of the theorem is equivalent to the

1542: assertion on the Smith normal form for the usual $3\times 3$ matrices over

1543: $\zet$.

1544: \ep

1545: 

1546: \begin{rem}

1547: {\rm

1548: An important feature of the theorem is that we have here a chain of

1549: embedded spaces

1550: \begin{equation}

1551: \h{\bin} \subset \h{\quat} \subset \h{\oct},

1552: \end{equation}

1553: and the action of the corresponding groups there.

1554: Theorem~\ref{np-orbits} gives us

1555: a uniform description of orbits for all three spaces

1556: in terms of diagonal matrices contained in each of these spaces.

1557: \ee

1558: }

1559: \end{rem}

1560: 

1561: 

1562: We also have the following trivial corollary to Theorem~\ref{np-orbits}

1563: (which also applies in the case of general $n\times n$ matrices over

1564: $\zet$ and their orbits under the elementary row and column transformations).

1565: 

1566: \begin{cor}

1567: Let $\h{\calg_\zet}$ and $G_\zet$ be as in Theorem~$\ref{np-orbits}$,

1568: and let $n$ be an integer, $n\ne 0$.

1569: The group $G_\zet$ acts transitively on set of matrices of determinant $n$

1570: if and only if $n$ is a squarefree integer.

1571: \end{cor}

1572: 

1573: 

1574: 

1575: \section{The Freudenthal construction}

1576: \label{sec:FC}

1577: 

1578: \subsection{Preliminary facts on the module $\mj$ and the group $\gr$}

1579: \label{ss:fc}

1580: 

1581: In this section we consider a certain class of modules and

1582: linear groups acting on them.

1583: Historically, the first example of this kind was introduced by

1584: H.~Freudenthal

1585: in \cite{fr14} in the process of constructing the $56$-dimensional

1586: representation of the group $E_7$ from the $27$-dimensional exceptional

1587: Jordan algebra.

1588: These modules were studied axiomatically (under the name of the Freudenthal

1589: triple systems) in \cite{b}, \cite{fau}, \cite{fer}.

1590: 

1591: We will consider examples of Freudenthal triple systems

1592: of the form $\mj$, where $\J$ is a cubic Jordan algebra

1593: of Section~$\ref{s:jordan}$.

1594: We will say that the module $\mj$ and its automorphism group $\gr$ (see below)

1595: are obtained from $\J$ using the~{\bf Freudenthal construction}.

1596: 

1597: 

1598: \medskip

1599: We will use \cite{b} as the main reference for this subsection.

1600: We consider a vector space $\m=\mj$ constructed from the space $\J$ in the

1601: following way

1602: \begin{equation}\label{eq:mj}

1603: \mj=F\oplus F \oplus \J \oplus \J,

1604: \qquad \mbox{ where $\J$ is a cubic Jordan algebra over $F$.}

1605: \end{equation}

1606: 

1607: We have $\dim \m=2\, \dim\J+2$, and an arbitrary element $x$ of the space

1608: $\m$ has the form

1609: \begin{equation} \label{gen}

1610: x=\mel\alpha \beta A B,\qquad

1611: {\rm where\ } \alpha, \beta \in F,\quad A,B \in \J.

1612: \end{equation}

1613: 

1614: We have the following skew-symmetric bilinear and quartic forms

1615: on $\m$ defined by

1616: $$

1617: \{ x,y\}= \alpha\delta - \beta\gamma+ (A,D)-(B,C)

1618: $$

1619: $$

1620: q(x)= 8 (A^\#, B^\#) -8\alpha \mnorm(A) - 8\beta \mnorm(B)

1621: -2\Bigl((A,B)-\alpha\beta\Bigr)^2.

1622: $$

1623: Here we have

1624: $x=\mel\alpha \beta A B$, $y=\mel\gamma \delta C D$.

1625: $(\cdot, \cdot)$ is the trace bilinear form,

1626: $\mnorm$ is the norm, and $\#$ is the

1627: sharp map in the Jordan algebra $\J$

1628: (see Subsections~$\ref{ss:sc}$,~$\ref{ss:ex}$ for detail).

1629: 

1630: Later on, when we turn to the integral case, it will be more convenient

1631: for us to consider the modified form $q'$:

1632: %

1633: \begin{equation}\label{qmod}

1634: q'(x)= -4 (A^\#, B^\#) +4\alpha \mnorm(A) +4\beta \mnorm(B)

1635: +\Bigl((A,B)-\alpha\beta\Bigr)^2,

1636: \qquad q=-2q'.

1637: \end{equation}

1638: 

1639: We will often refer to the form $q$ as the {\em norm form} or just the

1640: {\em norm} in the module~$\m$.

1641: 

1642: We can linearize the form $q$ and get a symmetric four-linear form

1643: $q(x,y,z,w)$ such that $q(x,x,x,x) = q(x)$. It follows that both the bilinear

1644: and four-linear form are non-degenerate. And hence we can define a

1645: trilinear operator $T:\m \times \m \times \m \to \m$ by the following

1646: rule: for given $x,y,z \in \m$, $T(x,y,z)$ is the unique element in $\m$

1647: such that

1648: \begin{equation} \label{deft}

1649: \{ T(x,y,z), w\} = q(x,y,z,w)\qquad \mbox{for any $w\in \m$}.

1650: \end{equation}

1651: 

1652: \begin{defn}

1653: The group $\gr$ is defined to be the group of all invertible $F$-linear

1654: transformations of $\m$ that preserve these forms, i.e.,

1655: \begin{equation} \label{invar}

1656: \Bigl\{ \sigma (x),\sigma (y)\Bigr\} = \{ x,y\},\qquad q\Bigl(\sigma (x)\Bigr) =

1657: q(x)\quad

1658: \end{equation}

1659: for any $\sigma \in \gr$.

1660: \end{defn}

1661: 

1662: It follows immediately from $(\ref{deft})$ and $(\ref{invar})$ that for

1663: any $\sigma \in \gr$

1664: \begin{equation} \label{autot}

1665: T\Bigl(\sigma (x), \sigma (y), \sigma (z)\Bigr)=

1666: \sigma \Bigl( T(x,y,z) \Bigr).

1667: \end{equation}

1668: 

1669: 

1670: We have the following four types of transformations in the group $\gr$:

1671: \begin{itemize}

1672: 

1673: \item[]

1674: For any $C\in \J$

1675: 

1676: \begin{equation} \label{phi}

1677: \phi(C)\quad : \quad \mel \alpha \beta A B\  \mto

1678: \mel {\alpha+(B,C)+(A, C^\#)+\beta \mnorm(C)} {\quad    \beta}

1679:  {\quad A+\beta C} {\quad  B+A\cross C +\beta C^\#}.

1680: \end{equation}

1681: 

1682: \item[]

1683: For any $D\in \J$

1684: 

1685: \begin{equation} \label{psi}

1686: \psi(D) \quad : \quad

1687: \mel \alpha \beta A B\  \mto

1688: \mel \alpha {\quad \beta + (A,D) + (B, D^\#) + \alpha \mnorm(D)}

1689: {A+ B\cross D+\alpha D^\#} {B+\alpha D}.

1690: \end{equation}

1691: 

1692: \item[]

1693: In addition, for every norm similarity $s\in \str\J$

1694: (cf. Definition~$\ref{def-groups}$)

1695: we have the transformation $\tf s \in \gr$ defined by:

1696: \begin{eqnarray} \label{t}

1697: \tf s& : &

1698: \mel \alpha\beta A B\  \mto

1699: \mel {\lambda ^{-1}\alpha}{\quad \lambda\beta}

1700:      {\quad s(A)}{\quad s^{*^{-1}}(B)}.

1701: \end{eqnarray}

1702: Here $\lambda \in \F$ is such that $\mnorm(s(x))=\lambda \mnorm(x)$,

1703: and $s^*$ is the linear transformation adjoint to $s$ with respect

1704: to the trace bilinear form $(\cdot, \cdot)$.

1705: We will mostly

1706: use such transformations when $s\in \npg\J$ and so $\lambda =1$.

1707: 

1708: \item[]

1709: Finally, we have the transformation $\tau$, which acts by:

1710: \begin{equation} \label{tau}

1711: \tau :  \quad

1712: \mel\alpha\beta A B \ \mto \ \mel {-\beta} {\alpha} {-B} A.

1713: \end{equation}

1714: \end{itemize}

1715: 

1716: The transformations $\phi(\cdot)$ and $\psi(\cdot)$ are conjugate to each

1717: other by $\tau$:

1718: $$

1719: \tau \phi(C) \tau^{-1}=\psi(-C).

1720: $$

1721: In addition we have the following relations in $\gr$:

1722: \begin{equation}\label{comm-rel}

1723: \begin{array}{rcl}

1724: \phi(-\one)\psi(\one)\phi(-\one) &=& \tau\\

1725: \tau^2 &=& -\id_\m\\

1726: \tau T(s) &=& T(s^{*-1})\tau\\

1727: T(s)\phi(C) &=& \phi(\lambda^{-1}\, s(C))\, T(s)\\

1728: T(s)\psi(C) &=& \psi(\lambda\, s^{*-1}(C))\, T(s).

1729: %

1730: %

1731: \end{array}

1732: \end{equation}

1733: Here $C\in\J$ and $s\in \str\J$ satisfies

1734: $\mnorm(s(D))=\lambda \mnorm (D)$.

1735: 

1736: 

1737: Note that when matrices $C$ and $D$ above have rank 1, the transformations

1738: $\phi$ and $\psi$ have the following simpler form:

1739: \begin{eqnarray}\label{t1s}

1740: \phi(C) & : &

1741: \mel\alpha\beta A B\  \mto\

1742: \mel{\alpha+(B,C)}{\quad    \beta}{\quad A+\beta C}{\quad  B+A\cross C}\\

1743:      \label{t2s}

1744: \psi(D) & : &

1745: \mel \alpha \beta A B\  \mto \

1746: \mel \alpha{\quad \beta + (A,D)}{\quad A+ B\cross D}{\quad    B+\alpha D}.

1747: \end{eqnarray}

1748: 

1749: \medskip

1750: The following proposition gives a more precise description of the group

1751: $\gr$ and its representation $\mj$. We will let $\astr\J$ (respectively,

1752: $\ag\m$) denote the algebraic group whose group of $F'$-rational points is

1753: $\astr{\J\tensor_F F'}$ (respectively, $\ag{\m\tensor_F F'}$) for all extension

1754: fields $F'$ of $F$. The symbol $\aJ$ will denote the vector (algebraic)

1755: group defined by $\J$, i.e., $\aJ(F')=\J\tensor_F F'$ taken with the

1756: additive group structure.

1757: 

1758: 

1759: 

1760: 

1761: \begin{prop}\label{prop:gener}

1762: \

1763: Let $\J$ be a Jordan algebra $\h{\calg}$ with $\calg=F,\bin, \quat, \oct$

1764: over a field $F$ (char$F=p\ne 2,3)$

1765: and let $\m=\mj$.

1766: 

1767: \begin{itemize}

1768: 

1769: \item[{\rm (i)}]

1770: 

1771: Then the group $\gr$

1772: is generated by elements $\phi (C)$, $\psi (D)$, $\tf s$.

1773: 

1774: \item[{\rm (ii)}]

1775: The  group $\gr$ is the set of $F$-points

1776: of an absolutely almost simple linear algebraic group $\ag\m$,

1777: which is defined over $F$ and $F$-split.

1778: 

1779: The group $\gr$ is connected (except the case $\J=\h\bin$). It has

1780: a two-element center, and its quotient modulo the

1781: center is a simple group of adjoint type.

1782: 

1783: In the case $\J=\h\bin$ the same result is true for the connected component

1784: $\ag\m\cc$, which is a subgroup of index $2$ in $\gr$.

1785: 

1786: \item[{\rm (iii)}]

1787: The following table lists the types of the group $\ag\m$ as well

1788: as the highest weight of its irreducible representation on the space $\m$.

1789: 

1790: $$

1791: \begin{array}{|c|c|c|}

1792: \hline

1793: \qquad \J \qquad       &  \mbox{Type of }\ag\m & \mbox{H.w. of }\ \m\\

1794: \hline

1795: \h{\F}    &  C_3 & \omega_3\\

1796: \h{\bin}  &  A_5 & \omega_3\\

1797: \h{\quat} &  D_6 & \omega_5\ {\rm or}\ \omega_6\\

1798: \h{\oct}  &  E_7 & \omega_7\\

1799: \hline

1800: \end{array}

1801: $$

1802: 

1803: \end{itemize}

1804: 

1805: \end{prop}

1806: 

1807: 

1808: {\sc Proof.}

1809: 

1810: 

1811: (i) was proved in \cite[Theorem~3]{b}. The theorem is stated there when $\J$

1812: is $27$-dimensional exceptional Jordan algebra, but it remains true for

1813: other algebras in the list. We will give a somewhat different proof

1814: of this statement when we describe the generators of the group $\gr$ in

1815: the integral case (Proposition~$\ref{genz}$).

1816: 

1817: \medskip

1818: (ii)

1819: It is not hard to see that algebraic equations defining the group $\ag\m$

1820: have integer coefficients.

1821: Then reducing modulo $p$, we can assume that it is defined over the prime

1822: field $F_p$, and hence over~$F$.

1823: 

1824: The $F$-split torus in $\ag\m$ is the image of the diagonal split torus of

1825: $\astr\J$ under the mapping

1826: $T:\astr\J\to\ag\m$, see (\ref{t}). Such a torus has the

1827: ``right rank", and it was described explicitly in Remark~\ref{groups}

1828: (the case of $\J=\h\oct$ was considered in \cite[Theorem 3.5]{gar}).

1829: 

1830: Next we notice that ${\phi(\aJ)}, {\psi(\aJ)}, T{(\astr\J)}$

1831: are closed subgroups

1832: of $\ag{\m}$, since they are homomorphic images of the algebraic

1833: groups $\aJ, \aJ, \astr\J$, respectively.

1834: These groups are connected when

1835: $\J=\F, \quat, \oct$ (see Remark~\ref{groups}).

1836: Since by (i) the group $\ag\m$ is generated by these subgroups,

1837: it is connected in these cases.

1838: 

1839: If $\J=\bin$, the group ${\astr\J}$ has two connected components, and in fact

1840: is isomorphic to the semi-direct product ${\astr\J}\cc\rtimes C_2$, where

1841: $C_2$ is the group of order $2$ generated by the element $t$

1842: corresponding to the transpose operation (Remark~\ref{groups}(ii)).

1843: This fact and commutation relations~$(\ref{comm-rel})$ imply that

1844: $\ag\m$ may have one or two components. The {\em algebraic} subgroup $\alg{H}$,

1845: generated by ${\phi(\aJ)}, {\psi(\aJ)}, T({\astr\J}\cc)$,

1846: is connected and closed. In addition, it is known

1847: (see, e.g., \cite[Section~$7.5$]{hum})

1848: %

1849: that $\alg H$ is

1850: generated by ${\phi(\aJ)}, {\psi(\aJ)}, T({\str\aJ}\cc)$ as an

1851: abstract group.

1852: The analysis of the action of the group $\gr$

1853: in the $20$-dimensional module $\mj$ (cf.

1854: Example~\ref{ex:sl6}) shows that the element $T(t)$ does not lie

1855: in $\alg H$.

1856: Hence we have a decomposition into two cosets

1857: $\ag\m=\alg H \cup T(t)\alg H$, which implies that $\alg H$ is the connected

1858: component of $\ag \m$.

1859: 

1860: The group $\gr$ contains the central element $\tau^2$, which acts as

1861: the scalar $-1$

1862: on the module $\m$. The statement about the center and the simplicity

1863: of the quotient modulo the center was proved in \cite[Theorem~6]{b}

1864: (the theorem was stated there for

1865: $\J=\h\oct$, but it remains true for other cases as well).

1866: %

1867: 

1868: \medskip

1869: (iii) If char$F=0$,

1870: the group $\ag\m$ may be identified by its Lie algebra, which is

1871: the Tits-Kantor-Koecher construction of $\J$ \cite{koe}.

1872: 

1873: An analysis of a slightly different version of the Freudenthal construction

1874: may be found in \cite[$2.22-2.26$]{sp-alg}

1875: with the resulting groups being identified in Section

1876: $14.31$ of the same book.

1877: 

1878: The corresponding irreducible representation can be identified by its

1879: highest weight vector and its dimension. The highest weight vector in such

1880: a representation may be chosen to be $(1,0,0,0)$ in $\mj$.

1881: \ep

1882: 

1883: \begin{rem} \label{rem:semis}

1884: {\rm

1885: Another example of an admissible cubic Jordan algebra

1886: $$

1887: \J=F\oplus Q_n, n\ge 1

1888: $$

1889: was given in Example~$\ref{ex:quad}$.

1890: 

1891: In this case the Freudenthal construction

1892: (assuming the quadratic form has maximal Witt index)

1893: produces the semi-simple $F$-split

1894: algebraic groups of type $\SL_2\times \SO_{n+2}$ acting on the

1895: tensor product $V_2 \tensor V(\omega_1)$ of the

1896: irreducible $\SL_2$-module $V_2$ of dimension two and the irreducible

1897: $\SO(n+2)$-module of highest weight $\omega_1$ (and dimension $n+2$).

1898: 

1899: For small $n$ the action of the group $\gr$ in $\m$ is essentially isomorphic

1900: to the action of the group $G$ in $\m$ listed in the table below.

1901: 

1902: $$

1903: \begin{array}{|l|l|l|l|}

1904: \hline

1905: n & G & \m & {\rm Comment}\\

1906: \hline

1907: 1 & \SL_2\times \SL_2 & V_2\tensor {\rm Sym}^2 V_2 &

1908: using\ Spin_3\cong \SL_2\\

1909: \hline

1910: 2 & (\SL_2)^3 & V_2\tensor V_2\tensor V_2 &

1911: using\ Spin_4\cong \SL_2\times \SL_2\\

1912: \hline

1913: 4 & \SL_2\times \SL_4 & V_2\tensor  \wedge^2(F^4) &

1914: using\ Spin_6\cong \SL_4\\

1915: \hline

1916: \end{array}

1917: $$

1918: 

1919: These cases are listed separately in Table $1$.

1920: }\ee

1921: \end{rem}

1922: 

1923: \begin{rem}\label{badred}

1924: {\rm

1925: We note that when char\,$F=2$, the quartic form $q'$ will reduce to

1926: $$

1927: q'(x)=(\alpha\beta-(A,B))^2,\qquad

1928: x\in \m.

1929: $$

1930: Hence the group $\gr$ will become the group of transformations preserving

1931: the symplectic form $\{\cdot, \cdot\}$ and the quadratic form

1932: $q'_2=(\alpha\beta-(A,B))$. Next we notice that the linearization of

1933: $q'_2$ will produce exactly the symplectic form $\{\cdot, \cdot\}$.

1934: Hence the group $\gr$ in the case char\,$F=2$ coincides with the orthogonal

1935: group on the vector space $\m$ with respect to the quadratic form $q'_2$.

1936: }\ee

1937: \end{rem}

1938: 

1939: We will conclude this subsection with an explicit description of the

1940: module $\m=\mj$ and the group $\gr$ for the Jordan algebra

1941: $\J=M_3(F)\cong \h\bin$.

1942: We will let $\gr\cc$ denote the subgroup generated by

1943: $\phi(\J), \psi(\J), T(\str\J\cc)$ for $\J=M_3(F)$

1944: (cf. Remark~$\ref{groups}$(ii)). This is a subgroup of index two in $\gr$,

1945: and it coincides with the group of $F$-rational points of the connected

1946: component of the algebraic group $\ag\m$ (cf. Proposition~$\ref{prop:gener}$).

1947: 

1948: \begin{example} \label{ex:sl6}

1949: {\rm

1950: We let $\J=M_3(F)$, and

1951: we are going to describe how the action of the group $\gr$

1952: in the $20$-dimensional module $\m=\mj$ is related to the action

1953: of the group $\SL_6(F)$ in the space $\wedge^3 F^6$.

1954: 

1955: Note that the group $\SL_6(F)$ has a center isomorphic to $\bfmu_6(F)$,

1956: and the quotient $\SL_6(F)/\bfmu_3(F)I_6$ acts faithfully in

1957: $\wedge^3 F^6$. We are going to explicitly describe an isomorphism of

1958: vector spaces $\theta: \mj\to \wedge^3 F^6$ and an isomorphism of groups

1959: $\theta':\gr\cc\to \SL_6(F)/(\bfmu_3(F)I_6)$ satisfying:

1960: \begin{equation}\label{ex:sl3iso}

1961: \theta(g\cdot v)=\theta' (g)\cdot \theta(v)\quad

1962: \mbox{ for }g\in \gr\cc, v\in\mj.

1963: \end{equation}

1964: With a slight abuse of notation we will still use $6\times 6$ matrices to

1965: represent elements of the quotient $\SL_6(F)/\bfmu_3(F)I_6$.

1966: 

1967: We begin by introducing the following notation. Let

1968: $$

1969: \{e_1, e_2, e_3, f_1, f_2, f_3\}

1970: $$

1971: be a standard basis of $F^6=F^3\oplus F^3$. Next we introduce the

1972: following ``dual" set of linearly independent vectors in $\wedge^2 F^6$:

1973: $$

1974: e_1^*=e_2\wedge e_3,\quad

1975: e_2^*=e_3\wedge e_1,\quad

1976: e_3^*=e_1\wedge e_2,

1977: $$

1978: $$

1979: f_1^*=f_2\wedge f_3,\quad

1980: f_2^*=f_3\wedge f_1,\quad

1981: f_3^*=f_1\wedge f_2.

1982: $$

1983: 

1984: Then we define a correspondence $\theta$ between bases of $\m$ and

1985: $\wedge^3F^6$ by

1986: \begin{eqnarray}\label{ex:sl1}

1987: \theta: \hspace{2cm}

1988: \mel 1000 & \mapsto & e_1\wedge e_2 \wedge e_3,\nonumber \\

1989: \mel 0100 & \mapsto & f_1\wedge f_2 \wedge f_3,\\

1990: \mel 00{E_{ij}}0 & \mapsto & e_i\wedge f_j^*,\nonumber\\

1991: \mel 000{E_{ij}} & \mapsto & f_i\wedge e_j^*,

1992: \hspace{4cm} 1\le i,j\le 3.

1993: \nonumber

1994: \end{eqnarray}

1995: 

1996: Next we define the homomorphism $\theta'$. First we will do it assuming

1997: that every element of $F$ is a cube of another element of $F$.

1998: The group $\gr\cc$ is generated by transformations $\phi(\J), \psi(\J),

1999: T(\str\J\cc)$.

2000: First we define the map $\theta'$ for $\phi(\J), \psi(\J)$ by

2001: \begin{equation}\label{ex:sl2}

2002: \theta':\quad

2003: \phi(A)\mto

2004: \maketwoblock   {I_3}{0}A{I_3},\qquad

2005: \psi(B)\mto

2006: \maketwoblock   {I_3}B{0}{I_3},

2007: \end{equation}

2008: where $A,B\in\J=M_3(F)$, and each block in

2009: $\maketwoblock\cdot\cdot\cdot\cdot$ represents

2010: a matrix in $M_3(F)$.

2011: 

2012: Next we define a homomorphism

2013: $$

2014: \theta'': GL_3(F)\times GL_3(F)\to \SL_6(F)/\bfmu_3(F)I_6

2015: $$

2016: by

2017: \begin{equation}\label{ex:sl3}

2018: \theta'':

2019: (A,B)\ \mapsto\

2020: \zeta_A\zeta_B

2021: \maketwoblock{A/\det (A)}0

2022: 0{B/\det (B)},\qquad

2023: A,B\in \GL_3(F), \zeta_A, \zeta_B\in F^\mult,

2024: \end{equation}

2025: where $\zeta_A, \zeta_B$ are chosen so that

2026: $$

2027: \zeta_A^3=\det(A), \quad

2028: \zeta_B^3=\det(B).

2029: $$

2030: 

2031: Since the map $\theta''$ has values in $\SL_6(F)/\bfmu_3(F)I_6$,

2032: the relation $(\ref{ex:sl3})$ does not depend on the choice of the third

2033: roots $\zeta_A, \zeta_B$. We also note that $\ker\theta''=F(I_3, I_3)$.

2034: Recall that $\str\J\cc \cong GL_3(F)\times GL_3(F)/F^\mult(I_3, I_3)$

2035: (cf. Remark~$\ref{groups}$(ii)),

2036: and hence the map $\theta''$ produces a well defined homomorphism

2037: \begin{equation}\label{ex:sl31}

2038: \theta': T(\str\J\cc) \to \SL_6(F)/\bfmu_3(F)I_6.

2039: \end{equation}

2040: 

2041: A direct computation shows that the generators of $\gr\cc$ and

2042: $\SL_6(F)/\bfmu_3(F)I_6$

2043: associated via $(\ref{ex:sl2}),(\ref{ex:sl3}), (\ref{ex:sl31})$

2044: define the

2045: same linear transformations in the (isomorphic) vector spaces

2046: $\m$ and $\wedge^3 \F^6$.

2047: Hence the maps $\theta, \theta'$ define an ``isomorphism" of the pairs

2048: $(\gr, \m)$ and $(\SL_6(F)/\bfmu_3(F)I_6,\ \wedge^3 F^6)$ in the sense

2049: of~$(\ref{ex:sl3iso})$.

2050: 

2051: In the case of an arbitrary field $F$, the quantities $\zeta_A,

2052: \zeta_B$ are elements of a suitable field extension of $F$.

2053: However, since $\zeta_A^3, \zeta_B^3$ are elements of $F$ and

2054: the matrix in~$(\ref{ex:sl3})$ acts in $\wedge^3 F^6$ (not just in $F^6$),

2055: the expression~$(\ref{ex:sl3})$ will produce a well defined $F$-linear

2056: transformation of $\wedge^3 F^6$.

2057: \ee

2058: }

2059: \end{example}

2060: 

2061: \subsection{The rank of elements of $\m$}

2062: \label{ss:rank}

2063: 

2064: \begin{defn}\label{def:rank}

2065: Let $\m=\m(\J)$, where $\J$ is a cubic Jordan algebra. The {\em

2066: rank} of an element $x\in \m$ is an integer between $0$ and $4$,

2067: which is uniquely defined by the following relations:

2068: \begin{itemize}

2069: \item

2070: rank $x = 4$  iff \ $q(x)\ne 0$;

2071: \item

2072: rank $x \le 3$  iff \ $q(x)=0$;

2073: \item

2074: rank $x \le 2$  iff \ $T(x,x,x)=0$;

2075: \item

2076: rank $x \le 1$  iff \  $3\,T(x,x,y)+\{x,y\}\,x=0$

2077: for all $y\in \m$;

2078: \item

2079: rank $x = 0$ iff \ $x=0$.

2080: \end{itemize}

2081: \end{defn}

2082: 

2083: The expressions defining rank $2,3,$ and $4$ are quite natural; they

2084: appeared elsewhere before, see, e.g., \cite{clerc}. %

2085: However our coordinate-free

2086: relation defining $\rank x\le 1$ appears to be new in this

2087: context.

2088: 

2089: The following lemma is an immediate corollary of the definitions of $q(x)$

2090: and $T(x,x,x)$.

2091: 

2092: \begin{lemma}\label{lem:rank}

2093: The rank of elements is preserved under the action of the group $\gr$.

2094: \end{lemma}

2095: 

2096: The expression defining the rank of $x$ are given by homogeneous

2097: algebraic equations in terms of the coordinates of $x$. We list

2098: the appropriate polynomials in the following

2099: 

2100: \begin{rem}\label{rem:rp}

2101: {\rm

2102: These polynomials will be described in terms of their value at an

2103: arbitrary element $x\in \m=\mj$ of the form

2104: \begin{equation}\label{elem1}

2105: x=\mel\alpha \beta{\ A}B.

2106: \end{equation}

2107: 

2108: \begin{itemize}

2109: \item

2110: The {\em quartic rank polynomial} is the quartic form $q$:

2111: \begin{equation} \label{p4}

2112: q(x)= 8 (A^\#, B^\#) -8\alpha \mnorm(A) - 8\beta \mnorm(B)

2113: -2[(A,B)-\alpha\beta]^2.

2114: \end{equation}

2115: 

2116: \item

2117: It was computed in \cite{b} that

2118: \begin{eqnarray}  \label{p3} \label{txxx}

2119: T(x,x,x)        \nonumber

2120: &=& \biggl( -\alpha ^2 \beta + \alpha\, (A,B)

2121: - 2 \mnorm(B), \qquad

2122: \alpha \beta ^2 -\beta\,(A,B) + 2 \mnorm(A),  \\

2123: && \phantom{-}

2124: 2B \cross A^\#-2\beta\, B^\# -[(A, B) - \alpha\beta]A,\\

2125: &&                 \nonumber -2A\cross B^\#+2\alpha\, A^\#

2126: +[(A,B)-\alpha\beta]B \biggr).

2127: \end{eqnarray}

2128: 

2129: \item

2130: We define {\em quadratic rank polynomials} to be the following expressions:

2131: \begin{eqnarray} \label{p2}

2132: \alpha A-B^\#, \quad \beta B - A^\# ,\qquad Q(x),

2133: \end{eqnarray}

2134: where $Q$ is a quadratic polynomial function with values in End$(\J)$,

2135: i.e., $Q(x)$ is a linear transformation of $\J$ defined by

2136: \begin{equation}\label{q}

2137: Q(x) : \quad C \quad \mto\quad (\alpha\beta - (A,B)) C +2\, V_{A,B}(C),

2138: \qquad \quad C\in \J.

2139: \end{equation}

2140: where $V_{A,B}$ was defined in $(\ref{a:v})$.

2141: \ee

2142: \end{itemize}

2143: }

2144: \end{rem}

2145: 

2146: \begin{lemma}\label{lem:rank1}

2147: Let $x=\mel\alpha\beta A B$ be an element of $\m=\m(\J)$.

2148: The quadratic rank polynomials

2149: \begin{equation} \label{e1r1}

2150: \alpha A-B^\#, \quad \beta B - A^\#,\quad Q(\bx) \qquad

2151: \mbox{ are equal to zero}

2152: \end{equation}

2153: if and only if

2154: \begin{equation} \label{e2r1}

2155: 3\,T(x,x,y)+\{x,y\}\,x \quad

2156: \mbox{ is equal to zero \ \ for all }y\in \m.

2157: \end{equation}

2158: \end{lemma}

2159: {\sc Proof.}

2160: It was computed in \cite[p.~88]{b} that linearization of $T(x,x,x)$ yields

2161: \begin{eqnarray*}  \label{p3lin}

2162: 3\,T(x,x,y)

2163: &=&

2164: \Bigl( -2\alpha \beta \gamma -\alpha^2 \delta + \gamma\, (A,B)

2165: +\alpha\, (C,B) +\alpha\, (A,D) - 6 \mnorm(B,B,D), \qquad\\

2166: && \quad\

2167: \gamma \beta ^2 +2\alpha\beta\delta-\delta\,(A,B)-\beta\, (C,B)

2168: -\beta\, (A,D)+ 6 \mnorm(A,A,C),  \\[2mm]

2169: &&\ \

2170: -[(C, B)+(A,D)-\alpha\delta - \beta\gamma]A

2171: -(A,B)C +\alpha\beta C-\\

2172: &&\ \ %

2173: - 2\delta B^\# -2\beta B\cross D

2174: +2 D \cross A^\#+2B\cross(A \cross C),\\[2mm]

2175: &&\quad\

2176: [(C,B)+(A,D)-\alpha\delta - \beta\gamma]B+

2177: (A,B)D -\alpha\beta D+\\

2178: &&\quad\

2179: +2\gamma A^\#+2\alpha A\cross C

2180: -2C\cross B^\#-2A\cross(B\cross D)

2181:  \Bigr)

2182: \end{eqnarray*}

2183: for $x=\mel\alpha \beta A B,\ y=\mel \gamma\delta C D$.

2184: 

2185: We can rewrite it using the identity ~$(\ref{a:id3})$

2186: and the definition of the form $\{\cdot, \cdot\}$ as

2187: \begin{equation} \label{r1eq}

2188: 3\,T(x,x,y)+\{x,y\}\,x =

2189: \end{equation}

2190: \vspace{-8mm}

2191: \begin{eqnarray*} \label{r1int-main}

2192: &=&

2193: \Bigl(

2194: -[3\alpha\beta-(A,B)]\gamma+2(\alpha A -B^\#, D),\\

2195: && \quad\

2196: [3\alpha\beta-(A,B)]\delta-2(\beta B -A^\#, C),\\

2197: && \quad\

2198: [3\alpha\beta-(A,B)]C- 2 (\beta B-A^\#)\cross D

2199: +2(\alpha A-B^\#)\delta-2Q(x)(C),\\

2200: && \ \,

2201: -[3\alpha\beta-(A,B)]D+ 2 (\alpha A-B^\#)\cross C

2202: -2(\beta B-A^\#)\gamma+2Q(x')(D)

2203: \Bigr).

2204: \end{eqnarray*}

2205: %

2206: 

2207: Here $x'=\mel\beta \alpha B A$.

2208: 

2209: \medskip

2210: It follow from the last expression (using that char\,$F\ne 2$) that

2211: $$

2212: 3\,T(x,x,y)+\{x,y\}\,x=0 \quad

2213: \mbox{for any }y\in \m

2214: $$

2215: 

2216: if and only if

2217: $$

2218: \alpha A-B^\#=0, \quad \beta B - A^\# =0,\quad

2219: Q(x)=0,\ Q(x')=0, \quad

2220: 3\alpha\beta-(A,B)=0.

2221: $$

2222: 

2223: This proves the ``$\Leftarrow$" implication of the lemma.

2224: 

2225: \medskip

2226: 

2227: Now suppose (\ref{e1r1}) holds. To prove ``$\Rightarrow$",

2228: we need to show that $Q(x')=0$ and $3\alpha\beta-(A,B)=0$.

2229: 

2230: We start with the latter one. We have $Q(x)(C)=0$ for any $C\in\J$.

2231: In particular, this is true for $C=\one$.

2232: We then compute using~(\ref{q}) and~$(\ref{traces})$:

2233: $$

2234: 0=\tr(Q(x)(\one))=3\alpha\beta-3(A,B)+2\tr(A\jprod  B)=

2235: 3\alpha\beta-3(A,B)+2(A, B)=3\alpha\beta-(A,B).

2236: $$

2237: 

2238: To prove $Q(x')=0$ we use the definition of the Jordan triple

2239: product~$(\ref{jtp})$ to notice that

2240: $$

2241: Q(x)(C)-Q(x')(C)=2Q(x)(\one)\jprod C,

2242: $$

2243: and the statement again follows from the fact that $Q(x)(C)=0$

2244: for any $C\in \J$.

2245: \ep

2246: 

2247: 

2248: \subsection{The canonical form in the case of a field}

2249: \label{ss:canf}

2250: 

2251: \begin{lemma} \label{bl-f}

2252: Let $\J$ be as in Proposition~$\ref{prop:gener}$ and $\m=\mj$.

2253: Every non-zero element of the module $\m$ can be brought to the form

2254: $$

2255: \mel 1 \beta {\ A} 0

2256: \qquad for\ some\ \beta\in \F,\ A\in\J

2257: $$

2258: by an appropriate element of $\gr$.

2259: \end{lemma}

2260: 

2261: {\sc Proof.} We start with an arbitrary element $x_1$ of the form

2262: $\mel {\alpha_1}{\beta_1} {A_1} {B_1}$.

2263: 

2264: First we show that we can transform it to an element in which the

2265: component $B_1$

2266: is not zero. If $B_1\ne 0$, there is nothing to do. If $A_1 \ne

2267: 0$, the transformation $\tau$ does the trick.

2268: %

2269: Now let us assume $A_1= B_1 = 0$ in $x_1$. After application of $\tau$ if

2270: necessary, we may assume that $\alpha _1\ne 0$. Then applying $\psi (D)$

2271: with any non-zero $D$, we get an element $x_2$ in which the

2272: component at the position $B_1$ is not zero.

2273: 

2274: 

2275: Thus we proceed with an element $x_2$ of the form $(\alpha _2,\beta_2,

2276: A_2, B_2)$ with $B _2\ne 0$. If $\alpha _2\ne 1$ we argue as follows.

2277: Since elements of $\J$ of rank $1$ span the whole $\J$, and the trace form

2278: in $\J$ is non-degenerate, we may find an element $C\in\J$ of rank 1, such

2279: that $(B_2,C)\ne 0$. Scaling $C$, we may assume that $\alpha_2+(B_2,C)=1$.

2280: The condition ${\rm rank}\,C=1$ implies $C^\# =0$ and $N(C)=0$. After

2281: applying the transformation $\phi (C)$ to $x_2$, we get an element $x_3$

2282: in which the first component is equal to $\alpha _2+(B_2,C)=1$.

2283: 

2284: We arrived at the element $x_3$ of the form

2285: $$

2286: \mel 1 {\beta_3}  {A_3} {B_3}.

2287: $$

2288: 

2289: The application of $\psi(D)$ with $D=-B_3$ brings this element to the

2290: desired form. \ep

2291: 

2292: \begin{lemma}

2293: \label{lcomp} An element

2294: $$

2295: \mel \alpha  \beta {\ \diag{a_1}{a_2}{a_3}} 0, \quad\alpha\ne 0

2296: $$

2297: can be mapped by a transformation in $\gr$ to the elements

2298: \begin{itemize}

2299: \item[\rm(i)]

2300: \hspace{2cm} $

2301: \mel \alpha {\ \beta-2\frac{a_2a_3}{\alpha}\,c}

2302: {\quad \diag{a_1+\beta c-\frac{a_2a_3}{\alpha}\, c^2}{\ a_2}{\ a_3}} 0,

2303: $

2304: \item[\rm(ii)]

2305: \hspace{2cm} $

2306: \mel \alpha {\ \beta-2\frac{a_1a_3}{\alpha}\,c}

2307: {\quad \diag{a_1}{\ a_2+\beta c-\frac{a_1a_3}{\alpha}\, c^2}{\ a_3}} 0, $

2308: \item[\rm(iii)]

2309: \hspace{2cm} $

2310: \mel \alpha {\ \beta-2\frac{a_1a_2}{\alpha}\,c}

2311: {\quad \diag{a_1}{\ a_2}{\ a_3+\beta c-\frac{a_1a_2}{\alpha}\, c^2}} 0, $

2312: \end{itemize}

2313: where $c$ is an arbitrary element in $\F$.

2314: 

2315: This lemma is also valid in the case $F=\zet$ assuming

2316: $\alpha|a_2, \alpha|a_3$ in {\rm (i)};

2317: $\alpha|a_1, \alpha|a_3$ in {\rm (ii)};

2318: $\alpha|a_1, \alpha|a_2$ in {\rm (iii)}.

2319: \end{lemma}

2320: 

2321: {\sc Proof.} The proof of this lemma is a direct computation. The action

2322: of

2323: $$

2324: \phi (C),\ C=\diag{c}00

2325: $$

2326: and then

2327: $$

2328: \psi(D),\ D=-\frac{1}{\alpha}\, \diag0{a_3c}{a_2c}

2329: $$

2330: gives the first element. The appropriate modifications of these yield the

2331: other two elements. \ep

2332: 

2333: \begin{rem} \label{rem1}

2334: {\rm

2335: When the component $B$ of an element of the form $(\ref{elem1})$ in $\m$ is

2336: equal to zero, the relations $(\ref{p4})-(\ref{p2})$ in the definition of

2337: the rank have the following simpler form

2338: \begin{itemize}

2339: \item

2340: rank $x \le 1$  iff

2341: \begin{equation} \label{r1s}

2342: \alpha A=0,\quad A^\# =0,\quad \alpha\beta=0;

2343: \end{equation}

2344: \vspace{-7mm}

2345: \item

2346: rank $x \le 2$  iff

2347: \begin{equation} \label{r2s}

2348: \alpha ^2\beta=0, \quad \alpha\beta ^2 +2N(A)=0, \quad \alpha\beta

2349: A=0,\quad \alpha A^\# =0;

2350: \end{equation}

2351: \vspace{-7mm}

2352: \item

2353: rank $x \le 3$  iff

2354: \begin{equation} \label{r3s}

2355: 8\alpha N(A)+2\alpha ^2\beta ^2=0\quad (i.e., q(x)=0);

2356: \end{equation}

2357: \vspace{-7mm}

2358: \item

2359: rank $x = 4$  iff

2360: \begin{equation} \label{r4s}

2361: q(x)\ne 0.

2362: \end{equation}

2363: \end{itemize}

2364: }\ee

2365: \end{rem}

2366: 

2367: \begin{theorem} \label{thm1}

2368: \

2369: 

2370: \begin{enumerate}

2371: \item[\rm (i)]

2372: {

2373: Let $F$ be a field of char\ $\ne 2,3$, let $\calg$ be the split composition algebra

2374: $\bin, \quat, \oct$ over $F$, and let $\J=\h\calg$.

2375: Let $(G, \m)$ be the pair (group, module) produced from $\J$

2376: by the Freudenthal construction.

2377: 

2378: Then

2379: 

2380: \begin{itemize}

2381: \item

2382: There exists a $G$-invariant quartic form (the norm) on the module $\m$.

2383: 

2384: \item

2385: The group $G$ acts

2386: transitively on the sets of elements of rank $1,2$, and $3$ in the module

2387: $\m$.

2388: 

2389: \item

2390: In the case of rank\/ $4$ the group $G$ acts transitively on the set of

2391: elements of a given norm $k$, for any $k\in F$, $k\ne 0$.

2392: \end{itemize}

2393: All these orbits are distinct, and the union of these orbits and $\{0\}$

2394: is the whole module~$\m$.

2395: }

2396: 

2397: \item[\rm (ii)]

2398: If in addition every element of $F$ is a square,

2399: then the same results apply to the pair $(G, \m)$ obtained

2400: from the Jordan algebra $\J=\h F$.

2401: \end{enumerate}

2402: This construction yields the simple algebraic groups and their irreducible

2403: representations described in Proposition~$\ref{prop:gener}$.

2404: \end{theorem}

2405: 

2406: {\sc Proof.}

2407: 

2408: The $G$-invariant quartic form $q$ defined on the space $\m$ is a part of

2409: the Freudenthal construction. One defines the rank of elements of $\m$ using the

2410: quartic form (Definition~$\ref{def:rank}$).

2411: 

2412: We are going to show that the group $\gr$ acts transitively on the set of

2413: elements of a given rank/given norm. Namely, we will show that every

2414: element of $\m$ is $\gr$-equivalent to one of the following elements

2415: in the ``canonical" form:

2416: \begin{eqnarray}

2417: \label{r1}

2418: \rank 1 & : & \mel 10 {\diag 000} 0,\\

2419: \label{r2}

2420: \rank 2 & : & \mel 10 {\diag 100} 0,\\

2421: \label{r3}

2422: \rank 3 & : & \mel 10 {\diag 110} 0,\\

2423: \label{r4}

2424: \rank 4 & : & \mel 10 {\diag 11k} 0, \quad k\in F, k\ne 0.

2425: \end{eqnarray}

2426: These elements have distinct rank/norm,

2427: and therefore by Lemma~$\ref{lem:rank}$ and the definition of $\gr$

2428: they lie in distinct orbits.

2429: 

2430: 

2431: We start with an arbitrary non-zero element $x$ in $\m$. By

2432: Lemma~\ref{bl-f} we can transform $x$ to the element

2433: \begin{equation} \label{semican}

2434: x_1=\mel 1 \beta {\ A} 0.

2435: \end{equation}

2436: We have rank\,$x$=rank\,$x_1$, and we can use Remark~\ref{rem1} when

2437: computing the rank of $x_1$. We also make use of the rank of elements

2438: of $\J$ (Definition~$\ref{def:jrank}$).

2439: \medskip

2440: 

2441: If rank\,$x_1=1$, then it follows from Remark~\ref{rem1} that $\beta=0$

2442: and $A=0$. So the element $x_1$ is already in the form~$(\ref{r1})$.

2443: 

2444: If rank\,$x_1=2$, then it follows from Remark~\ref{rem1} that $\beta=0$.

2445: The condition $\alpha A^\# =0$ of~$(\ref{r2s})$ implies rank\,$A=1$.

2446: By Proposition~$\ref{prop:np-o}$

2447: there exists $s\in \npg\J$ such that

2448: $$

2449: s(A)=\diag{1}00.

2450: $$

2451: The action of the transformation $\tf{s}\in \gr$ on the element $x_1$

2452: brings it to the form~$(\ref{r2})$ as desired.

2453: \medskip

2454: 

2455: Let us now consider the case of rank\,$x_1=3$. We start by showing that we

2456: can assume that the $\beta$-component of $x_1$ is zero. If $\beta\ne 0$,

2457: it follows from $(\ref{r3s})$ that

2458: $$

2459: N(A)=-\frac{1}{4}\beta ^2\ne 0.

2460: $$

2461: Again, by Proposition~$\ref{prop:np-o}$

2462: %

2463: we can find an element $s\in \npg\J$ that brings $A$ to the diagonal

2464: form:

2465: $$

2466: s(A)=\diag{1}1n\quad {\rm with\ } n=N(A).

2467: $$

2468: And hence we have

2469: $$

2470: \tf{s}(x_1)=\mel 1 \beta {\ \diag{1}1n} 0.

2471: $$

2472: Taking $c=\beta/2$ in Lemma~\ref{lcomp}(iii) we see that we can map

2473: %

2474: the above element to the element

2475: $$

2476: x_2=\mel 1 0 {\ A'} 0

2477: $$

2478: with component $\beta$ equal to zero.

2479: 

2480: We still have $\rank x_2=3$, and

2481: it follows from $(\ref{r3s})$ that $N(A')=0$. On the other hand, $A^\#

2482: \ne 0$, since the opposite would imply that rank\,$x_2 \le 2$. Hence

2483: rank\,$A'=2$. Again we can find an $s'\in \npg\J$ that brings $A'$ to the

2484: diagonal form:

2485: $$

2486: s'(A')=\diag{1}10,

2487: $$

2488: and hence

2489: $$

2490: \tf{s'}(A')=\mel 1 0 {\ \diag{1}10} 0.

2491: $$

2492: This is an element of the form~$(\ref{r3})$,

2493: and we are done with the case of rank 3.

2494: 

2495: \medskip

2496: The last case is the case of rank $4$.

2497: 

2498: We are still working with an element $x_1$ of the form $(\ref{semican})$.

2499: Since $\rank x_1=4$, we have $q(x_1)\ne 0$.

2500: 

2501: There are three possible subcases for the element

2502: $$

2503: x_1=\mel 1 \beta {\ A} 0.

2504: $$

2505: 

2506: \medskip

2507: {\em Subcase $1$.} $\beta =0$.

2508: 

2509: In this subcase the expression for the norm becomes

2510: $$

2511: q(x_1)=-8\, N( A).

2512: $$

2513: Since $q(x_1)\ne 0$, we have $N(A)\ne 0$, and we can find an $s\in \npg\J$

2514: that brings $A$ to the form $\diag{1}1k$, $k\ne 0$. It follows that

2515: $\tf{s}(x_1)$ has the form~$(\ref{r4})$ as desired.

2516: 

2517: \medskip

2518: {\em Subcase $2$.} $\beta \ne 0$ and rank\,$A\ge 2$.

2519: 

2520: We start by bringing $A$ to the diagonal form of Proposition~$\ref{prop:np-o}$

2521: by an appropriate

2522: element $s\in \npg\J$. Then $s(A)$ has the form $\diag{1}1n$, $n\in \F$. We

2523: can apply Lemma~\ref{lcomp}(iii) with $c=\beta /2$ to the element

2524: $\tf{s}(x_1)$, and get a new element with $\beta$-component equal to zero.

2525: This brings us to the subcase 1, and we are done.

2526: 

2527: \medskip

2528: {\em Subcase $3$.} $\beta \ne 0$ and rank\,$A\le 1$.

2529: 

2530: We again start by bringing $A$ to the diagonal form by an

2531: appropriate element $s\in \npg\J$. Then $s(A)$ has the form

2532: $\diag{\epsilon}00$, $\epsilon=0$ or $1$ depending on the rank\,$A$. We

2533: can apply Lemma~\ref{lcomp}(ii) and (i) (if necessary) with $c=1/\beta$,

2534: and obtain element

2535: $$

2536: x_2=\mel \alpha \beta {\diag 110} 0.

2537: $$ This brings us to

2538: the subcase 2, and we are done again.

2539: 

2540: We showed that an arbitrary non-zero element can be brought to one of the

2541: elements in the canonical form~$(\ref{r1})-(\ref{r4})$.

2542: Hence the union of the orbits described in the hypothesis

2543: is the whole~$\m$.

2544: \ep

2545: 

2546: \section{The Integral version of the Freudenthal construction}

2547: \subsection{The integral structure in the module $\mj$}

2548: \label{ss:intstr}

2549: 

2550: In this section we consider the integral structure $\mz$ in the module

2551: $\m$ and the integral form $\grz$ of the group $\gr$ that acts on it.

2552: We define

2553: \begin{equation}\label{mz}

2554: \mz\defeq\m(\jz)=\zet\oplus\zet\oplus\jz\oplus\jz,

2555: \end{equation}

2556: where $\jz$ is one of

2557: \begin{equation} \label{zlist}

2558: \jz:\qquad

2559: \h{\octz},\  \h{\quatz},\  \h{\binz},\

2560: \h{\zet},\  \zet\oplus \zet \oplus \zet,

2561: \end{equation}

2562: see Subsection~$\ref{ss:33int}$

2563: 

2564: Most of the definitions and assertions of this section apply also in the

2565: cases

2566: $$

2567: \zet\ \mbox{ and }\ \zet\oplus Q_n(\zet) \quad

2568: \mbox{(cf.

2569: Remark~\ref{rem:semis})}.

2570: $$

2571: However these cases require somewhat different treatment and

2572: we will not consider them in this paper.

2573: 

2574: 

2575: An arbitrary element of the module $\mz$ has the form

2576: $$

2577: \mel\alpha\beta AB,

2578: \qquad

2579: \mbox{where } \alpha, \beta \in \zet \mbox{ and } A,B \in \jz.

2580: $$

2581: 

2582: Since $\jz$ is closed under $\#$, and $\tr$ and $\btr(\cdot, \cdot)$ are

2583: integer-valued,

2584: we have that the quartic form $q$, the modified form $q'$

2585: (see \ref{qmod})), and the skew-symmetric form~$\{\cdot,\cdot\}$ have integer

2586: values on $\mz$. The expression~$(\ref{txxx})$ implies

2587: that $T(x,x,x)\in \mz$ for $x\in\mz$.

2588: 

2589: \begin{defn}

2590: {\rm

2591: We define the {group $\grz$} to be the group of invertible

2592: $\zet$-linear transformations of $\mz$ that preserve the quartic

2593: form $q$ and the bilinear form $\{\cdot, \cdot \}$.

2594: }

2595: \end{defn}

2596: It is immediate that transformations

2597: \begin{equation} \label{elz}

2598: \phi(C), \psi(D),\ \tf{s},\ \tau \quad

2599: {\rm with}\ C,D\in \jz,\ s\in \str{\jz}

2600: \end{equation}

2601: preserve $\mz$, and therefore lie in the group $\grz$.

2602: 

2603: \begin{defn}

2604: {\rm

2605: We will call the transformations of the form $(\ref{elz})$ the

2606: {\em elementary\/} transformations of $\mz$.

2607: }

2608: \end{defn}

2609: 

2610: If $F$ is the field of characteristic zero, then

2611: the group $\grz$ may also be defined as the set of integral points

2612: of the group $\gr$ with respect to the integral structure on the

2613: module $\m$ determined by (\ref{mz}).

2614: 

2615: We will prove in Proposition~$\ref{genz}$ below that $\grz$ is generated by

2616: the elementary transformations~$(\ref{elz})$.

2617: We will make use of this assertion in the next examples, which give an

2618: explicit description of $\mz$ and $\grz$ in two special cases.

2619: 

2620: \begin{example} \label{ex:sl6int}

2621: {\rm

2622: This example gives a  description of the integral version

2623: of Example~$\ref{ex:sl6}$.

2624: Here we have

2625: $\jz=M_3(\zet)\cong\h\binz$ and

2626: $\mz=\m(\jz)=\zet\oplus\zet\oplus\jz\oplus\jz$.

2627: By analogy with Example~$\ref{ex:sl6}$, we define the group $\grz\cc$ to

2628: be the subgroup of $\grz$ generated by

2629: $\phi(\jz)$, $\psi(\jz)$, $T(\str\jz\cc)$.

2630: We are going to establish the isomorphism of the integral pairs

2631: $(\grz\cc, \jz)\cong(\SL_6(\zet), \wedge^3\zet^6)$ in the sense

2632: of~$(\ref{ex:sl3iso})$

2633: 

2634: We keep the notation for bases and define the isomorphism $\theta$ of

2635: $\zet$-modules $\mz$ and $\wedge^3 \zet^6$ as in Example~$\ref{ex:sl6}$.

2636: 

2637: We have $\bfmu_3(\zet)=\{1\}$ and hence

2638: $\SL_6(\zet)=\SL_6(\zet)/\bfmu_3(\zet)I_6$ acts faithfully in

2639: $\wedge^3 \zet^6$.

2640: We are going to define the map

2641: $\theta':\grz\cc\to\SL_6(\zet)$ in a way similar to Example~$\ref{ex:sl6}$.

2642: 

2643: 

2644: First we define $\theta'$ for $\phi(\jz), \psi(\jz)$ by

2645: \begin{equation}\label{ex:sl2int}

2646: \theta':\quad

2647: \phi(A)\mto

2648: \maketwoblock   {I_3}{0}A{I_3},\qquad

2649: \psi(B)\mto

2650: \maketwoblock   {I_3}B{0}{I_3},

2651: \end{equation}

2652: where $A,B\in\jz=M_3(\zet)$.

2653: 

2654: As for the group $\str\jz\cc$, we notice that

2655: it consists of the transformations of the form

2656: $$

2657: X\mapsto AXB^{-1}

2658: $$

2659: with $X\in\jz, A,B\in \GL_3(\zet)$ and $(A,B)$ acting trivially iff

2660: $A=B=\pm I_3$ (cf. Remark~$\ref{groups}$(ii)).

2661: 

2662: Next we will define a homomorphism

2663: $$

2664: \theta'': \GL_3(\zet)\times \GL_3(\zet)\to \SL_6(\zet).

2665: $$

2666: Since in this case the determinants of the matrices involved are $\pm 1$,

2667: we can define $\theta''$ by the following simpler formula:

2668: \begin{equation}\label{ex:sl3int}

2669: \theta'':

2670: (A,B)\ \mapsto\ \maketwoblock{(\det B)A}0

2671: 0{(\det A)B},\qquad

2672: A,B\in \GL_3(\zet).

2673: \end{equation}

2674: 

2675: The map $\theta''$ factors through the quotient

2676: $\GL_3(\zet)\times \GL_3(\zet)/\{\pm(I_3, I_3)\}$ and yields a well-defined

2677: map

2678: \begin{equation}\label{ex:sl31int}

2679: \theta': T(\str\jz\cc)\to \SL_6(\zet).

2680: \end{equation}

2681: 

2682: A direct computation shows that the generators of the groups $\grz\cc$ and

2683: $\SL_6(\zet)$

2684: associated via $(\ref{ex:sl2int}),(\ref{ex:sl3int}), (\ref{ex:sl31int})$

2685: define the

2686: same linear transformations in the (isomorphic) $\zet$-modules

2687: $\mz$ and $\wedge^3 \zet^6$.

2688: Hence the maps $\theta, \theta'$ define an ``isomorphism" of the pairs

2689: $(\grz\cc, \mz)$ and $(\SL_6(\zet),\ \wedge^3 \zet^6)$ as desired.

2690: 

2691: It will be proved in Proposition~$\ref{genz}$ that every element in $\grz$

2692: is a product of the elementary transformations~$(\ref{elz})$. Commutation

2693: relations~$(\ref{comm-rel})$ imply that every element in $\grz$ has

2694: the form $g_0$ or $T(t) g_0$, where $g_0\in \grz\cc$ and $t$ is the

2695: transpose operation, see Remark~$\ref{groups}$(ii).

2696: Hence we have $\grz=\grz\cc \cup T(t)\grz\cc$.

2697: \ee

2698: }

2699: \end{example}

2700: 

2701: \begin{example}\label{ex:FFF}

2702: {\rm

2703: Let $\J=F\oplus F \oplus F$ and hence $\jz=\zet\oplus\zet\oplus\zet$,

2704: $\mz=\m(\jz)=\zet\oplus \zet\oplus \jz\oplus \jz$.

2705: In this example we are going to explicitly describe the relation between

2706: the action of the group $\grz$ in the $8$-dimensional module $\mz$ and the

2707: action of the group $\SL_2(\zet)\times \SL_2(\zet)\times \SL_2(\zet)$ in the space

2708: $\zet^2\tensor \zet^2 \tensor \zet^2$.

2709: 

2710: We will use the more compact notation $\Gamma$ to denote the

2711: group $\SL_2(\zet)\times \SL_2(\zet)\times \SL_2(\zet)$.

2712: Note that the group

2713: $\Gamma$

2714: has a center that consists of the matrices $\{(\pm I_2, \pm I_2, \pm I_2)\}$,

2715: and the quotient

2716: $\Gamma/K_4$ acts faithfully in

2717: $\zet^2\tensor \zet^2 \tensor \zet^2$, where

2718: $$

2719: K_4=\{(I_2, I_2, I_2),(I_2, -I_2, -I_2),(-I_2, I_2, -I_2),(-I_2, -I_2, I_2),\}.

2720: $$

2721: 

2722: 

2723: Let $G_0$ be the subgroup in $\grz$ generated by transformations

2724: $\phi(\jz)$, $\psi(\jz)$, and $T(\pm \id_\jz)$.

2725: We are going to describe explicitly an isomorphism of $\zet$-modules

2726: $\theta:\mz \to \zet^2\tensor\zet^2\tensor\zet^2$ and an isomorphism

2727: of groups

2728: $\theta':G_0 \to\Gamma/K_4$ satisfying

2729: \begin{equation}\label{ex:F0}

2730: \theta(g\cdot v)=\theta' (g)\cdot \theta(v)\quad

2731: \mbox{ for }g\in G_0, v\in\mz.

2732: \end{equation}

2733: With a slight abuse of notation we will use triples of $2\times 2$

2734: matrices to

2735: represent elements of the quotient $\Gamma/K_4$.

2736: 

2737: 

2738: First we will establish an isomorphism of $\zet$-modules

2739: $\mz\cong\zet^2\tensor\zet^2\tensor\zet^2$.

2740: 

2741: 

2742: 

2743: Let $\{E_1, E_2, E_3\}$

2744: be the standard basis of $\jz=\zet\oplus\zet\oplus\zet$,

2745: and let $\{e_1, e_2\}$ be the standard basis of $\zet^2$.

2746: We define $\theta$ on bases and extend it by $\zet$-linearity

2747: \begin{eqnarray} \label{ex:F1}

2748: &&

2749: \mel 1000 \mapsto e_1\tensor e_1\tensor e_1,\quad\ \

2750: \mel 0100 \mapsto e_2\tensor e_2\tensor e_2,\nonumber \\

2751: &&

2752: \mel 00{E_1}0 \mapsto e_1\tensor e_2\tensor e_2,\quad

2753: \mel 00{E_2}0 \mapsto e_2\tensor e_1\tensor e_2,\quad

2754: \mel 00{E_3}0 \mapsto e_2\tensor e_2\tensor e_1,\quad\\

2755: &&

2756: \mel 000{E_1} \mapsto e_2\tensor e_1\tensor e_1,\quad

2757: \mel 000{E_2} \mapsto e_1\tensor e_2\tensor e_1,\quad

2758: \mel 000{E_3} \mapsto e_1\tensor e_1\tensor e_2.\nonumber

2759: \end{eqnarray}

2760: 

2761: Next we define the map $\theta'$ on generators of $G_0$

2762: (and onto generators of $\Gamma/K_4$) by

2763: \begin{eqnarray} \label{ex:F2}

2764: \phi(a_1, a_2, a_3) & \mapsto&

2765: \Biggl(

2766: {%

2767: \maketwo{1}{a_1}{0}{1},

2768: \maketwo{1}{a_2}{0}{1},

2769: \maketwo{1}{a_3}{0}{1}}

2770: \Biggr)\nonumber \\

2771: \psi(a_1, a_2, a_3) & \mapsto &

2772: \Biggl(

2773: \maketwo{1}{0}{a_1}{1},\maketwo{1}{0}{a_2}{1},\maketwo{1}{0}{a_3}{1}

2774: \Biggr)\\

2775: T(\pm \id_\jz) & \mapsto &

2776: \pm

2777: \Bigl(

2778: I_2, I_2, I_2

2779: \Bigr),\nonumber

2780: \end{eqnarray}

2781: where $(a_1, a_2, a_3)\in \jz$.

2782: 

2783: A direct computation shows that the generators of $G_0$ and

2784: $\Gamma/K_4$

2785: associated via $(\ref{ex:F2})$

2786: define the

2787: same linear transformations in the (isomorphic) $\zet$-modules

2788: $\mz$ and $\zet^2\tensor \zet^2\tensor \zet^2$.

2789: Hence the maps $\theta, \theta'$ define an ``isomorphism" of the pairs

2790: $(G_0, \mz)$ and $(\Gamma/K_4,\ \zet^2\tensor \zet^2\tensor \zet^2)$

2791: in the sense of~$(\ref{ex:F0})$.

2792: }\ee

2793: \end{example}

2794: 

2795: 

2796: \subsection{Reduction in the module $\m(\jz)$}

2797: \label{ss:red}

2798: 

2799: \begin{defn}

2800: {\rm

2801: Given a free $\zet$-module $M$,

2802: we say that an integer $d$ {\em divides} an

2803: element $x\in M$ if $x=dx'$ for some $x'\in M$.

2804: 

2805: We define the $\gcd$ of a collection of elements in

2806: $M$ to be the greatest integer that

2807: divides all these elements. By definition the $\gcd$ is  a positive

2808: integer, if at least one of the elements is nonzero.

2809: }

2810: \end{defn}

2811: 

2812: \begin{defn}\label{prim}

2813: {\rm

2814: An element $x$ of a free $\zet$-module $M$ is said to be {\em primitive},

2815: if $\gcd (x)=1$.

2816: }

2817: \end{defn}

2818: Evidently, for any non-zero element $x$ we have

2819: $

2820: x = \gcd(x)\ x',

2821: $

2822: where $x'$ is primitive.

2823: 

2824: \begin{lemma} \label{gcd}

2825: Let $M$ be a free $\zet$-module and

2826: let $g$ be an element in {\rm End}${}_\zet(M)$ such that $g^{-1}$ exists and

2827: also lies in {\rm End}${}_\zet(M)$.

2828: Then for any $x$ in $M$ and any non-zero integer\/ $d$

2829: $$

2830: d\ divides\ x\qquad

2831: \mbox{\rm if and only if}\qquad

2832: d\ divides\ g(x).

2833: $$

2834: \end{lemma}

2835: 

2836: {\sc Proof.}

2837: Obvious.

2838: 

2839: \bigskip

2840: 

2841: %

2842: %

2843: %

2844: %

2845: 

2846: \begin{defn}\label{def:reduced}

2847: {\rm

2848: An element $x$ of the module $\mz=\m (\jz)$ is said to be {\em reduced},

2849: if it is of the form

2850: \begin{equation} \label{reduced}

2851: x=\mel\alpha\beta{\ A}0,\qquad

2852: \alpha,\beta\in \zet,\ A\in\jz,

2853: \end{equation}

2854: with $\alpha>0,\quad \alpha |\beta, \ \alpha |A$. We say that $x$ is a

2855: {\em diagonal reduced} element, if in addition $A$ is a diagonal matrix.

2856: }

2857: \end{defn}

2858: 

2859: We note that for a reduced element $x$ as in (\ref{reduced}) we have

2860: $\gcd(x)=\alpha$. And a reduced $x$ is primitive if and only if

2861: $\alpha=1$.

2862: 

2863: 

2864: 

2865: %

2866: %

2867: 

2868: \begin{lemma}[\sc Reduction Lemma I]  \label{bl-fz}

2869: \

2870: 

2871: \hspace{-\parindent}%

2872: Let $\jz$ be one of $\h{\binz}, \h{\quatz}, \h{\octz}, \zet\oplus\zet\oplus\zet$.

2873: Every non-zero element ${\bx}$ of the module $\mz=\m (\jz)$

2874: is equivalent to a {\em diagonal reduced} element under the action of

2875: a series of elementary transformations

2876: \begin{equation} \label{elem-tr}

2877: \phi (C), \psi(D), \tf{s}, \tau \quad

2878: with \ C, D \in \jz,\ s\in \npg{\jz}.

2879: \end{equation}

2880: \end{lemma}

2881: 

2882: {\sc Proof.}

2883: 

2884: {\bf (a)}

2885: First we are going to prove the assertion when $\jz$

2886: is one of $\h{\binz}, \h{\quatz}, \h{\octz}$.

2887: 

2888: We are going to apply transformations $\phi (C), \psi(D), \tf{s}$

2889: to obtain a reduction

2890: procedure similar to the Gaussian elimination process for

2891: regular integer matrices.

2892: 

2893: The proof consists of two parts.

2894: We describe four reduction steps in the first part, and in the second part

2895: we show how we use these steps to bring an arbitrary non-zero element to a

2896: reduced form.

2897: 

2898: {\bf Part I.}

2899: 

2900: We start the proof of the lemma by explaining how we can use the

2901: elementary transformations~(\ref{elem-tr}) to get the reduction.

2902: We will use the

2903: relations $(\ref{phi})$, $(\ref{psi})$, and the ability to bring any

2904: element in $\jz$ to the (diagonal) Smith normal form by an element

2905: in $\npg \jz$ (see Theorem~\ref{np-orbits}).

2906: Note that transformations $s\in \npg\jz$ preserve the norm, and hence

2907: we have $\lambda=1$ in (\ref{t}).

2908: 

2909: The four reduction steps are summarized below:

2910: \medskip

2911: 

2912: (RED1): \

2913: $\mel\alpha\beta{\ A}B \longrightarrow \mel{g_1}**0$

2914: with $0<g_1\le\ \min \{|\alpha|,\gcd( B)\}$

2915: \medskip

2916: 

2917: (RED2): \

2918: Assume $\alpha>0$, \ $\alpha \not | \gcd(\beta, A)$;

2919: 

2920: \ \phantom{(RED2): }%

2921: $\mel\alpha\beta{\ A}0 \longrightarrow \mel{*}{\beta'}{A'}{{}*}$

2922: with $0<\min\{\beta', \gcd(A')\}<\alpha$

2923: 

2924: \bigskip

2925: (RED3): \

2926: $\mel\alpha\beta{\ A}B \longrightarrow \mel *{g_2}0{{}*}$

2927: with $0<g_2\le \min \{|\beta|,\gcd(A)\}$

2928: 

2929: \medskip

2930: (RED4): \

2931: Assume $\beta>0$, \ $\beta \not | \gcd(\alpha, B)$;

2932: 

2933: \ \phantom{(RED4): }%

2934: $\mel\alpha \beta{\ 0} B \longrightarrow \mel{\alpha'} *{*} {B'}$

2935: with $0<\min\{\alpha', \gcd(B')\}<\beta$

2936: 

2937: 

2938: \medskip

2939: The symbol ``$*$" above represents some element at the appropriate

2940: position. We will not be interested in that entry for the moment.

2941: 

2942: The idea of (RED1) is to use the interaction of the elements $(\alpha, B)$

2943: in the left column to replace them by $(g_1, 0)$. (RED2) is designed to be

2944: applied after (RED1), and its idea is to (eventually) reduce the

2945: pair $(\beta, A)$ in the right column modulo $g_1$.

2946: 

2947: (RED3) and (RED4) are mirror images of (RED1) and (RED2) with the roles of

2948: the left and right columns reversed.

2949: 

2950: We now proceed to describing these steps in detail.

2951: \medskip

2952: 

2953: (RED1): \

2954: $\mel\alpha\beta{\ A} B \longrightarrow \mel{g_1} * * 0$

2955: with $0<g_1\le\ \min \{|\alpha|,\gcd( B)\}$

2956: 

2957: We start with an element $x\in \mz$ of the form

2958: \begin{equation}

2959: x=\mel\alpha\beta{\ A} B,\qquad

2960: \alpha,\beta\in \zet,\ A,B\in\jz.

2961: \end{equation}

2962: 

2963: We can use the transformation $\psi (D)$ with an appropriately chosen $D \in

2964: \jz$ to reduce entries of the element $B$ modulo $\alpha$. This

2965: essentially means that the $9, 15,$ or $27$ integers in the entry $B+\alpha D$ of

2966: $\psi(D)({\bx})$ will lie in the interval $0 \dots |\alpha | -1$.

2967: 

2968: Conversely, we can reduce the entry $\alpha$ modulo $m_2$, where

2969: $m_2 = \gcd (B)$.

2970: Without loss of generality we can assume that $B$ is in the Smith normal

2971: form (to get it, first find an $s\in \npg\jz$ such that $s(B)$ is in the

2972: Smith canonical form; then replace $x$ by $\tf{s^{*^{-1}}}\ ({\bx}))$;

2973: the value of $\alpha$ will not be affected by this). This remark

2974: implies that the component $B$ of the (new) element $x$ will have the form

2975: $$

2976: B=\diag{m_2}**\qquad{\rm with}\ m_2=\gcd(B).

2977: $$

2978: Next, we find

2979: an integer $c_1$ such that $\alpha +m_2 c_1$ is between $1$ and $m_2$,

2980: and let $C$ be the diagonal matrix $\diag{c_1}00$.

2981: The ``$1,1$"-entry of the element $\phi (C) ({\bx})$ is

2982: $$

2983: \alpha+(B,C)+(A, C^\# ) +\beta \mnorm(C)

2984: $$

2985: and since rank $C$=1, it is equal to

2986: $$

2987: \alpha+(B,C)=\alpha+m_2 c_1,

2988: $$

2989: which is in $\{1,\dots, m_2\}$,

2990: and we have reduced $\alpha$ modulo $\gcd(B)$

2991: (note that we chose the set of residues,

2992: which excludes $0$, but contains $m_2$, since we want to keep this

2993: component non-zero).

2994: 

2995: We can now reduce the entries $\alpha$ and $B$ modulo each other, and

2996: continue to do so as long as the entry at the position $B$ is non-zero.

2997: The value of the component $\alpha$ will decrease after each pair of

2998: iterations, while remaining positive. This condition guarantees that the

2999: process will terminate at some step. This procedure

3000: will bring $x$ to the desired form, and since at each step we reduced

3001: modulo $\alpha$ or $\gcd(B)$, the condition on $g_1$ is satisfied. This

3002: completes the description of (RED1).

3003: 

3004: \bigskip

3005: 

3006: (RED2): \

3007: Assume $\alpha>0$, \ $\alpha \not | \gcd(\beta, A)$;

3008: 

3009: \ \phantom{(RED2): }%

3010: $\mel \alpha\beta A0 \longrightarrow \mel *{\beta'}{A'}{{}*}$

3011: with $0<\min\{\beta', \gcd(A')\}<\alpha$.

3012: 

3013: We start with an element $x$ as stated in the assumption.

3014: We can apply a transformation $\tf{s}$ to bring the entry $A$ to the

3015: Smith normal form without changing $\alpha$ and $\beta$, so without

3016: loss of generality we assume that $A$ is already in the Smith normal form

3017: $A=\diag{a_1}{a_2}{a_3}$, $\ a_1|a_2,\ a_2|a_3,\ \alpha>0$.

3018: 

3019: Let us now take matrix $D=\diag{0}1{d_3}$ (we will choose the value for

3020: $d_3$ later). Then, since $B=0$ and $\mnorm(D)=0$, we have

3021: $$

3022: \psi(D)({\bx})=

3023: \mel\alpha {\ \beta +(A,D)}{\quad  A+\alpha D^\# }{\ \alpha D}.

3024: $$

3025: 

3026: We have

3027: $D^\# =\diag{d_3}00$, and hence the ``$1,2$"-entry of the above element

3028: is

3029: $$

3030: \diag{a_1+d_3 \alpha}{\quad a_2}{\quad a_3}.

3031: $$

3032: We can find $d_3\in \zet$ that brings $a'_1=a_1+d_3 \alpha$ into the

3033: integer interval $1,\dots,\alpha$.

3034: If $a'_1<\alpha$, then we are done. And if $a'_1=\alpha$, it means

3035: $\alpha | A$. Then

3036: we can use the element $a'_1$, ideas from (RED3), and the non-divisibility

3037: assumption (which translates now into $\alpha \not| \beta$),

3038: to bring the entry in the positions $\beta$

3039: to the interval $1,\dots,\alpha -1$.

3040: This completes the description of (RED2).

3041: 

3042: The steps (RED3) and (RED4) are mirror images of (RED1) and (RED2).

3043: They are performed in a similar way.

3044: 

3045: \medskip

3046: 

3047: {\bf Part 2.}

3048: 

3049: Now we show how we can use the above reduction steps

3050: to get the $\gcd ({\bx})$

3051: of a non-zero element

3052: $$

3053: {\bx}=\mel\alpha\beta A B,

3054: $$

3055: 

3056: at the position $\alpha$ or $\beta$.

3057: 

3058: In the loop described below

3059: we will apply our reductions to the appropriate matrix, but we will

3060: concentrate our attention on either the pair in the right column or

3061: the pair in the left column of the matrix $\mel\alpha\beta AB$.

3062: 

3063: Without loss of generality we can assume that either of $\alpha$ or $B$ is

3064: non-zero (otherwise we apply $\tau$ and the requirement is satisfied).

3065: \medskip

3066: 

3067: \hspace{10mm}

3068: \begin{minipage}{135mm}

3069: \renewcommand{\baselinestretch}{1}

3070: \renewcommand{\arraystretch}{1}

3071: \normalsize

3072: 

3073: {\sc Loop.}

3074: 

3075: {\em Step 1.}

3076: We have an element of the form

3077: $$

3078: \mel{\alpha'} {\beta'} {A'} {B'}

3079: $$

3080: with non-zero pair $\alpha', B'$ in the left column.

3081: 

3082: By (RED1) we can bring it to the form

3083: $$

3084: \mel {g_1}\beta A 0

3085: $$

3086: with some (new) values of $\beta, A$ and

3087: $0<g_1\le \min\{|\alpha'|, \gcd (B')\}$. It follows from (\ref{end2})

3088: below that $g_1$ is strictly smaller than

3089: the value of $g_2$ from the previous step

3090: ({\em ignore this remark if you just entered the loop}).

3091: 

3092: We need to consider two cases:

3093: \begin{itemize}

3094: 

3095: \item

3096: $g_1 | \beta, g_1 |A$

3097: 

3098: We are done: leave the loop.

3099: 

3100: \item

3101: $g_1$ does not divide either $\beta$ or $A$

3102: 

3103: It follows that $g_1 \not | \gcd (\beta, A)$.

3104: The assumptions of (RED2) are now satisfied, and we apply it to

3105: $\mel {g_1} \beta A 0$. The result is a new element of the form

3106: $\mel *{\beta''}{A''} {{}*}$ with

3107: \begin{equation} \label{end1}

3108: 0<\min\{ \beta'',\gcd (A'')\}<g_1.

3109: \end{equation}

3110: 

3111: 

3112: \end{itemize}

3113: Proceed to {\em Step 2}.

3114: %

3115: \vspace{3mm}

3116: 

3117: %

3118: %

3119: \end{minipage}

3120: 

3121: \hspace{10mm}

3122: \begin{minipage}{135mm}

3123: \renewcommand{\baselinestretch}{1}

3124: \renewcommand{\arraystretch}{1}

3125: \normalsize

3126: 

3127: 

3128: {\em Step 2.}

3129: 

3130: This step is a mirror image of the previous step, with the roles of elements in the

3131: left column and the right column exchanged.

3132: 

3133: We have an element of the form

3134: $$

3135: \mel{\alpha''} {\beta''} {A''} {B''}

3136: $$

3137: with elements in the right column satisfying (\ref{end1}).

3138: 

3139: Using (RED3) we can bring it to the form

3140: $$

3141: \mel\alpha{g_2} 0 B

3142: $$

3143: with some (new) values of $\alpha, B$ and

3144: $0<g_2\le \min\{|\beta''|,\ \gcd(A'')\}$. And (\ref{end1}) implies that~

3145: $g_2$ is strictly smaller than the value of $g_1$ from the previous step.

3146: 

3147: We consider two cases again:

3148: \begin{itemize}

3149: 

3150: \item

3151: $g_2 |\alpha, g_2 | \gcd (B)$

3152: 

3153: We are done: leave the loop.

3154: 

3155: \item

3156: $g_2$ does not divide either $\alpha$ or $B$

3157: 

3158: 

3159: It follows that $g_2 {}\not\!|{\ } \gcd (\alpha, B)$. %

3160: The assumptions of (RED4) are now satisfied, and we apply it to

3161: $\mel \alpha {g_2} 0 B$. The result is a new element of the form

3162: $\mel {\alpha'} **{B'}$ with

3163: \begin{equation} \label{end2}

3164: 0<\min\{ \alpha',\gcd (B')\}<g_2.

3165: \end{equation}

3166: \end{itemize}

3167: Proceed to {\em Step 1}.

3168: \end{minipage}

3169: 

3170: \vspace{4mm}

3171: As we run this loop the values of $g_1$ and $g_2$ will decrease

3172: at each even and odd step, respectively.

3173: On the other hand, they both remain positive

3174: integers, and this guarantees us that we leave the loop at some step.

3175: 

3176: If we left the loop at the first step, then the first entry of the

3177: resulting element is $\gcd({\bx})$. And if we left it at the second

3178: step, then the $\gcd({\bx})$ is at the second position. In the latter

3179: case we apply $\tau^{-1}$ to bring it to the first entry.

3180: 

3181: Let us change the notation again (we have done it many times recently),

3182: and denote the obtained element by

3183: $$

3184: \mel\alpha\beta A0

3185: $$

3186: (it follows from the above that the ``$2,1$"-entry is equal to zero).

3187: 

3188: We can apply a transformation from $\npg\jz$,

3189: if necessary, to convert $A$ to the Smith normal form. Then $A$

3190: turns into a diagonal matrix, and the last requirement of the lemma is

3191: satisfied.

3192: 

3193: This element has the desired form.

3194: \bigskip

3195: 

3196: {\bf (b)}

3197: We now consider the case $\jz=\zet\oplus\zet\oplus\zet$

3198: (and hence $\mz=\m(\zet\oplus\zet\oplus\zet)$).

3199: In this case our ``matrices" in $\jz$ are already in the ``diagonal" form.

3200: But the norm-preserving group is too small in this case,

3201: and it is not true that every element of $\jz$ may be converted to a Smith

3202: normal form. However, one can still use the ideas from the algorithm in part (a)

3203: with minor modifications to prove the assertion of the lemma.

3204: 

3205: We will not pursue the route described in the previous paragraph to prove~(b).

3206: Instead we will use the relation between the pairs

3207: $\Bigl(\grz, \mz\Bigr)$ and

3208: $\Bigl(\SL_2(\zet)\times \SL_2(\zet)\times \SL_2(\zet),\

3209: \zet^2\tensor \zet^2\tensor \zet^2\Bigr)$ described in Example~$\ref{ex:FFF}$.

3210: 

3211: The reduction procedure described in \cite[Appendix]{bh1}

3212: explains how an arbitrary element of $\zet^2\tensor \zet^2\tensor \zet^2$

3213: may be transformed to a certain $5$-parameter form by a transformation in

3214: the group $\SL_2(\zet)\times \SL_2(\zet)\times \SL_2(\zet)$.

3215: Using the isomorphisms of Example~$\ref{ex:FFF}$,

3216: this procedure may be restated in terms of

3217: transformations in $\grz$ acting on the elements on $\m(\jz)$,

3218: which yields the desired result.

3219: \ep

3220: 

3221: An immediate corollary of this reduction lemma is Proposition~$\ref{genz}$

3222: that describes the generators of the group $\grz$.

3223: 

3224: \begin{prop} \label{genz}

3225: Let $\jz$ be one of $\h{\binz}, \h{\quatz}, \h{\octz},

3226: \zet\oplus\zet\oplus\zet$, and let

3227: $\mz=\m(\jz)$.

3228: The group $\grz$ is generated by the elementary transformations

3229: \begin{equation} \label{ele}

3230: \phi(C),\ \psi(D),\ \tf{s},\ \tau

3231: \end{equation}

3232: with $C,D\in \jz,\ s\in \str\jz$.

3233: \end{prop}

3234: 

3235: {\sc Proof.}

3236: 

3237: Let $\rho$ be an arbitrary element in the group $\grz$. We need to prove

3238: that $\rho$ can be represented as a product of elementary transformations

3239: $(\ref{ele})$.

3240: 

3241: Let $f_1$ be

3242: the element $\mel 1000\in \mz$. Let us consider the element

3243: $$

3244: \bx \defeq \rho (f_1).

3245: $$

3246: 

3247: By Lemma~$\ref{bl-fz}$ we can use a product $\eta$ of elementary

3248: transformations to bring $\bx$ to the (diagonal) reduced form

3249: $$

3250: \bx'=\mel\alpha\beta A 0,\qquad \alpha>0.

3251: $$

3252: 

3253: %

3254: %

3255: %

3256: %

3257: %

3258: We have $\bx' = \eta\, \rho (f_1)$, and it follows from %

3259: Lemma~$\ref{lem:rank}$ that

3260: $$

3261: {\rm rank\ }\bx'={\rm rank\ }f_1=1.

3262: $$

3263: 

3264: Then it follows from the relations in Remark~$\ref{rem1}$

3265: that $\beta = 0$, $A=0$, and hence

3266: $$

3267: \bx'=\mel{\alpha}000.

3268: $$

3269: 

3270: We have $\eta, \rho\in \grz$, and it follows from Lemma~$\ref{gcd}$ that

3271: $\gcd(\bx')=\gcd(f_1)$,

3272: and hence we get that

3273: $$

3274: \alpha=1 \qquad {\rm and} \qquad \bx'=f_1,

3275: $$

3276: 

3277: and hence

3278: $$

3279: \eta \rho (f_1)=f_1.

3280: $$

3281: 

3282: Repeating the argument of \cite[Lemma 12,13]{b} and using the fact

3283: that our transformations preserve the integral structure,

3284: we prove that $\eta \rho$ has the form

3285: $$

3286: \eta \rho = \tf{s}\phi(C) \quad

3287: {\rm for\ some \ }s\in \str\jz, C\in \jz,

3288: $$

3289: and since $\eta$ was the product of elementary transformations, $\rho$ has

3290: the desired form.

3291: \ep

3292: 

3293: 

3294: 

3295: 

3296: \subsection{Projective elements in the module $\m(\jz)$}

3297: \label{ss:proj}

3298: 

3299: 

3300: In this subsection we discuss the concept of projective elements,

3301: which plays the central role in the classification of orbits in the

3302: integral case.

3303: 

3304: \begin{defn}\label{def:proja}{\bf (a)}

3305: 

3306: {\rm

3307: Let $\jz=

3308: \zet\oplus\zet\oplus\zet$, and let $\mz=\m(\jz)$.

3309: An element $x\in \mz$

3310: $$

3311: x=

3312: \mel\alpha\beta

3313: {({a_1}, {a_2}, {a_3})}

3314: {({b_1}, {b_2}, {b_3})},\qquad

3315: \alpha, \beta, a_i, b_i\in \zet

3316: $$

3317: is said to be {\em projective}, if each of the three binary quadratic forms

3318: associated to this element

3319: \begin{eqnarray*}

3320: \biggl(\alpha a_1-b_2 b_3\biggr)x^2 +

3321: \biggl(-a_1b_1+a_2 b_2+ a_3 b_3 -\alpha\beta\biggr)xy +

3322: \biggl(\beta b_1-a_2a_3\biggr) y^2,\\[2mm]

3323: \biggl(\alpha a_2-b_3 b_1\biggr)x^2 +

3324: \biggl(\phantom{-}a_1b_1-a_2 b_2+ a_3 b_3 -\alpha\beta\biggr)xy +

3325: \biggl(\beta b_2-a_3a_1\biggr) y^2,\\[2mm]

3326: \biggl(\alpha a_3-b_1 b_2\biggr)x^2 +

3327: \biggl(\phantom{-}a_1b_1+a_2 b_2- a_3 b_3 -\alpha\beta\biggr)xy +

3328: \biggl(\beta b_3-a_1a_2\biggr) y^2.

3329: \end{eqnarray*}

3330: is primitive, i.e., the $\gcd$ of its coefficients is equal to one.

3331: 

3332: }

3333: \end{defn}

3334: 

3335: The concept of projective element was first introduced by M.~Bhargava

3336: \cite{bh-phd, bh1} for elements of

3337: $\zet^2\tensor\zet^2\tensor\zet^2$. The equivalence of the two definitions

3338: follows from the isomorphism

3339: $$

3340: \zet^2\tensor\zet^2\tensor\zet^2 \cong \m(\zet\oplus\zet\oplus\zet)

3341: $$

3342: established in Example~\ref{ex:FFF}.

3343: 

3344: The concept of projective element was extended to

3345: $\wedge^3\zet^6$ (and several more examples) in~\cite{bh1}.

3346: More precisely, M.~Bhargava showed that there is a natural injection

3347: \begin{equation}\label{inc1}

3348: \zet^2\tensor\zet^2\tensor\zet^2 \hookrightarrow \wedge^3\zet^6

3349: \end{equation}

3350: and in addition every element of $\wedge^3\zet^6$ is

3351: $\SL_6(\zet)$-equivalent to an element in the image of

3352: $\zet^2\tensor\zet^2\tensor\zet^2$ under the embedding (\ref{inc1}).

3353: An element of $\wedge^3\zet^6$ was defined to be {\em projective},

3354: if its $\SL_6$-orbit contains an image under (\ref{inc1}) of a

3355: projective element in $\zet^2\tensor\zet^2\tensor\zet^2$.

3356: 

3357: In our case we have a chain of natural inclusions of cubic vector spaces

3358: ($\zet$-modules):

3359: $$

3360: \zet\oplus\zet\oplus\zet \subset

3361: \h\binz \subset \h\quatz \subset \h\octz

3362: $$

3363: (the first inclusion being the diagonal embedding), which induces the

3364: natural chain

3365: \begin{equation} \label{chain3}

3366: \m(\zet\oplus\zet\oplus\zet) \subset

3367: \m(\h\binz) \subset \m(\h\quatz) \subset \m(\h\octz).

3368: \end{equation}

3369: 

3370: The first two modules in the last chain are isomorphic to

3371: $\zet^2\tensor\zet^2\tensor\zet^2$ and $\wedge^3\zet^6$, respectively

3372: (see Examples~$\ref{ex:FFF}, \ref{ex:sl6int}$). We proved in Lemma~\ref{bl-fz}

3373: that an arbitrary element of

3374: each of the modules $\mz$ in (\ref{chain3}) is $\grz$-equivalent to

3375: a {\em diagonal reduced} element,

3376: which can be thought of as an element of the smallest submodule

3377: $\m(\zet\oplus\zet\oplus\zet)$ in the chain~$(\ref{chain3})$.

3378: This lemma allows us to extend

3379: the concept of a projective element to $\m(\h\binz)\cong \wedge^3\zet^6$,

3380: using an argument different from Bhargava's, and further extend this

3381: concept to $\m(\h\quatz)$ and  $\m(\h\octz)$.

3382: 

3383: %

3384: \hspace{-\parindent}{\bf Definition~\ref{def:proja} (b)}

3385: {\rm

3386: Let $\jz$ be one of $\h{\binz}, \h{\quatz}, \h{\octz}$, and let

3387: $\mz=\m(\jz)$. An element $x\in\mz$ is said to be {\em projective}, if its

3388: $\grz$-orbit contains a diagonal reduced element (cf. Lemma~$\ref{bl-fz}$)

3389: \begin{equation}\label{proj-el}

3390: \mel\alpha\beta{\diag{a_1} {a_2} {a_3}}{0},\qquad

3391: \alpha, \beta, a_i\in \zet,\quad

3392: \alpha>0,\quad \alpha |\beta, \ \alpha |a_i,

3393: \end{equation}

3394: which is projective.

3395: }

3396: %

3397: 

3398: We note that for an element of the form (\ref{proj-el}) the projectivity

3399: conditions of Definition~\ref{def:proja}(a) become

3400: \begin{eqnarray} \label{def:projexpr}

3401: & \gcd(\alpha a_1, \alpha\beta, a_2a_3)=1 & \nonumber\\

3402: & \gcd(\alpha a_2, \alpha\beta, a_1a_3)=1 &\\

3403: & \gcd(\alpha a_3, \alpha\beta, a_1a_2)=1 & \nonumber

3404: \end{eqnarray}

3405: Taking into account the divisibility conditions of reduced elements

3406: (Definition~\ref{def:reduced}), it follows that if element (\ref{proj-el})

3407: is projective, then $\alpha=1$, i.e., $x$ is primitive. Nevertheless we

3408: kept $\alpha$ in the expressions (\ref{def:projexpr}) to emphasize the fact

3409: that they are homogeneous expressions of degree $2$.

3410: \medskip

3411: 

3412: The definition of a projective element we just gave is not quite

3413: satisfactory, since

3414: it is not convenient to work with orbit representatives of the group

3415: $\grz$, which may be quite large.

3416: Next we would like to argue that the concept of a projective

3417: element is related to certain equations of degree $3$, which will allow us

3418: to get a simple projectivity test in {\em almost} all cases.

3419: 

3420: \begin{prop} \label{prop:proj}

3421: {\rm

3422: Let $\jz$ be as in Definition~$\ref{def:proja}$(a,b) and $\mz=\m(\jz)$.

3423: Let $x$

3424: be in $\mz$ and let $T(x,x,x)$ be defined as in~$(\ref{deft})$.

3425: \begin{enumerate}[(i)]

3426: \item

3427: If $\gcd\, T(x,x,x)=1$ then $x$ is projective;

3428: \item

3429: If $\gcd\, T(x,x,x)\ge 3$ or $T(x,x,x)=0$ then $x$ is not projective.

3430: \item

3431: When $q'(x)$ is odd, $\gcd T(x,x,x)=1$ iff $x$ is projective.

3432: \end{enumerate}

3433: }

3434: \end{prop}

3435: 

3436: {\sc Proof.}

3437: 

3438: First we notice that the quantity $\gcd T(x,x,x)$ is invariant with

3439: respect to the action of the group $\grz$.

3440: This assertion follows from the

3441: $\gr$-invariance of the quartic form $q(x)$ and the symplectic form

3442: $\{\cdot,\cdot\}$,

3443: the definition of $T(x,x,x)$ $(\ref{deft})$, and Lemma~$\ref{gcd}$

3444: (see Lemma~$\ref{di}$ below for a detailed argument).

3445: Hence, for a diagonal reduced element

3446: \begin{equation}\label{x1a}

3447: x_1=\mel\alpha\beta{\diag{a_1} {a_2} {a_3}}{0},\qquad

3448: \alpha, \beta, a_i\in \zet,\quad

3449: \alpha>0, \alpha|\beta, \alpha| a_i.

3450: \end{equation}

3451: contained in the $\grz$-orbit of $x$ by Lemma~$\ref{bl-fz}$ we have

3452: $$

3453: \gcd T(x,x,x)=\gcd T(x_1,x_1,x_1).

3454: $$

3455: 

3456: It is more convenient to work with $x_1$,

3457: since the components

3458: of $T(x_1,x_1, x_1)$ have the following simple form (see~$(\ref{txxx})$):

3459: $$

3460: T(x_1,x_1,x_1)=

3461: \mel{\alpha^2\beta}{\ \alpha\beta^2+2\mnorm(A)}{\ \alpha\beta A}{\ 2\alpha

3462: A^\#},

3463: $$

3464: 

3465: where $A=\diag{a_1} {a_2} {a_3}$,

3466: $\mnorm (A)=a_1a_2a_3,\ A^\#=\diag{a_2a_3}{a_3a_1}{a_1a_2}$.

3467: 

3468: 

3469: {\bf (i)}

3470: Let $\gcd T(x_1,x_1,x_1)=1$, and suppose $x_1$ is not projective. It means

3471: that the $\gcd$ of at least one of the expressions in~$(\ref{def:projexpr})$

3472: is greater than $1$. Without loss of generality we assume that there

3473: exists a prime $p>1$ such that

3474: $$

3475: p\ \Bigl|\  \alpha a_3,\  \alpha\beta,\ a_1a_2.

3476: $$

3477: This implies that $p$ divides each component of $T(x_1, x_1, x_1)$,

3478: contrary to our assumption.

3479: 

3480: {\bf (ii)}

3481: We are given $x$ such that $\gcd T(x, x, x)\ge 3$, and need to show that

3482: $x$ is not projective. Let $x_1$ be {\em any} diagonal reduced element of

3483: the form~$(\ref{x1a})$ in the $\grz$-orbit of $x$. Hence

3484: $$

3485: \gcd T(x_1, x_1, x_1)=

3486: \gcd

3487: \mel{\alpha^2\beta}{\alpha\beta^2+2\mnorm(A)}{\alpha\beta A}{2\alpha A^\#}

3488: \ge 3.

3489: $$

3490: 

3491: First we notice that if $\alpha >1$, then $x_1$ is not primitive.

3492: Non-primitivity implies non-projectivity, and hence there is nothing to

3493: prove in this case.

3494: Thus from now on we may assume

3495: $\alpha=1$, and hence

3496: $$

3497: T(x_1, x_1, x_1)=

3498: \mel{\beta}{\beta^2+2\mnorm(A)}{\beta A}{2 A^\#}.

3499: $$

3500: 

3501: Suppose that there exists a prime $p\ge 3$ that divides $\gcd T(x_1,

3502: x_1, x_1)$. Then we have $p|\beta$. In addition, the condition $p|A^\#$

3503: implies that at least two of the numbers $a_i$ are divisible by $p$. Then

3504: $p$ divides at least one of the expressions~$(\ref{def:projexpr})$, and

3505: hence $x_1$ is not projective.

3506: 

3507: If the prime $p$ such as in the previous paragraph does not exist, it

3508: follows from the conditions that $\gcd T(x_1, x_1, x_1)$ is a power of $2$

3509: (and is greater than $2$). Then we have $4|\beta$, $2|A^\#$, and repeating

3510: the argument of the previous paragraph, we get that $2$ divides at least

3511: one of the expressions~$(\ref{def:projexpr})$. Hence $x_1$ is not

3512: projective in this case either.

3513: 

3514: We have proved that if $\gcd T(x,x, x)\ge 3$, then every diagonal reduced

3515: element in $\grz$-orbit of $x$ is not projective, and hence $x$ is not

3516: projective.

3517: 

3518: This argument also applies in the case $T(x,x,x)=0$, if one replaces the

3519: condition ``divisible by $p$" with the condition ``equal to $0$".

3520: 

3521: {\bf (iii)}

3522: The implication ``$\Rightarrow$" was proved in (i). To prove the converse

3523: implication, we again look at the diagonal reduced projective

3524: $x_1$ of the form $(\ref{x1a})$ in the $\grz$-orbit of $x$.

3525: 

3526: It follows from (ii) that $\gcd T(x_1,x_1,x_1)$ is equal to $1$ or $2$,

3527: and we will show that the latter option is not possible.

3528: 

3529: Since $x_1$ is projective, it is primitive, and hence $\alpha=1$.

3530: We then have $q'(x_1)=\beta^2+4\mnorm (A)$, and since $q'(x)=q'(x_1)$ is

3531: odd, we have that $\beta$ is odd. This conclusion implies that $2$ cannot

3532: divide $\gcd T(x_1,x_1,x_1)$, and hence $\gcd T(x_1,x_1,x_1)=\gcd T(x,x,x)=1$.

3533: \ep

3534: 

3535: \begin{rem}

3536: {\rm

3537: The above proposition  provides a projectivity test for all cases

3538: except $\gcd\,$ %

3539: $T(x,x,x)=2$. Elements

3540: \begin{equation}

3541: x_1=\mel{1}{2}{\diag{1}12}0 \qquad

3542: x_2=\mel{1}{2}{\diag{1}22}0

3543: \end{equation}

3544: both have $\gcd \Bigl(T(x_i,x_i,x_i)\Bigr)=2$.

3545: However $x_1$ is projective and $x_2$ is not projective in the sense of

3546: Definition~$\ref{def:proja}(a)$.

3547: }

3548: \end{rem}

3549: 

3550: \begin{cor}\label{cor:anyrep}

3551: {\rm

3552: Proposition~$\ref{prop:proj}$ implies that if $\gcd T(x,x,x)$ is

3553: different from $2$ then the (non)proje\-ctivity of $x$ is determined by

3554: {\em any} representative of its $\grz$-orbit

3555: in $\m(\zet\oplus\zet\oplus\zet)$.

3556: }

3557: \end{cor}

3558: 

3559: It appears that the assertion of Corollary~$\ref{cor:anyrep}$

3560: remains true for $x$ with $\gcd\, T(x,x,x)=2$,

3561: however the quantity $T(x,x,x)$ does not seem to be precise enough to

3562: treat this case. Apparently, the difficulty of proving this fact

3563: is related to the ``bad" reduction

3564: of the quartic form in the case when char\,$F=2$ (cf. Remark~$\ref{badred}$).

3565: \medskip

3566: 

3567: Our next remark is concerned with one more characterization of projective

3568: elements.

3569: There is a natural choice of a basis and coordinates in the module

3570: $\mz=\m(\jz)$ with $\jz=\zet\oplus\zet\oplus\zet,\h\binz,\h\quatz,\h\octz$.

3571: In each of these cases the quartic form $q'(x)$ may be viewed as a

3572: homogeneous polynomial in $n$ variables, $n=8, 20, 32, 56$, respectively.

3573: And the components of $T(x,x,x)$ are nothing else but the $n$ partial

3574: derivatives of the polynomial $q'(x)$ (divided by $2$, since all

3575: these partial derivatives have even coefficients).

3576: In other words, each component of the formal gradient vector

3577: $\frac{1}{2}\nabla q'(x)$ is a polynomial with integer coefficients,

3578: and these expressions are the same as the components of $T(x,x,x)$.

3579: These observations and Proposition~$\ref{prop:proj}$ yield the following

3580: 

3581: \begin{cor}\label{cor:nabla}

3582: {\rm

3583: Let $x$ be in $\mz=\m(\jz)$ with $\jz$ as in Definition~$\ref{def:proja}$(a,b).

3584: Let $q'$ be the quartic invariant of the module $\mz$, see~$(\ref{qmod})$.

3585: Then $\gcd T(x,x,x)=\gcd \frac{1}{2}\nabla q'(x)$ and hence

3586: \begin{itemize}

3587: \item

3588: If $\gcd\, \frac{1}{2}\nabla q'(x)=1$ then $x$ is projective;

3589: \item

3590: If $\Bigl(\gcd\, \frac{1}{2}\nabla q'(x)\ge 3\Bigr)$

3591: or $\Bigl(\nabla q'(x)=0\Bigr)$ then $x$ is not projective.

3592: \end{itemize}

3593: }

3594: \end{cor}

3595: 

3596: 

3597: 

3598: \begin{rem}\label{rem:nabla}

3599: {\rm

3600: It follows

3601: from general considerations that if $p(x)$ is a (quartic) form on a

3602: $\zet$-module $V_\zet$ invariant with respect to a group $G_\zet$, then

3603: $\gcd \nabla p(x)$ is invariant with respect to $G_\zet$. This observation

3604: suggests that Corollary~$\ref{cor:nabla}$ may be used to describe

3605: (non)projective elements in spaces ${\rm Sym}^3\zet^2,

3606: \zet^2\tensor{\rm Sym}^2\zet^2, \zet^2\tensor\zet^4$

3607: underlying higher composition laws

3608: related to quadratic rings (see Table~1 and \cite{bh1} for details).

3609: }

3610: \end{rem}

3611: 

3612: 

3613: \subsection{Further reduction and the classification of the projective orbits}

3614: \label{ss:fred}

3615: 

3616: \begin{lemma}[\sc Reduction Lemma II]  \label{red2}

3617: \

3618: 

3619: \hspace{-\parindent}%

3620: Let $\jz$ be one of $\h{\binz}, \h{\quatz}, \h{\octz}$

3621: and $\mz=\m(\jz)$.

3622: Let ${\bx}\in \mz$ be a projective element.

3623: 

3624: 

3625: Then $x$ is equivalent to an element

3626: \begin{equation} \label{proj1}

3627: \mel 1 \varepsilon{\ \diag{1}1k} 0,\qquad

3628: \varepsilon\in\{0,1\},\quad

3629: k\in \zet

3630: \end{equation}

3631: 

3632: under a series of elementary transformations

3633: \begin{equation} \label{elem-tr1}

3634: \phi (C), \psi(D), \tf{s}, \tau \quad

3635: with \ C, D \in \jz,\ s\in \npg{\jz}.

3636: \end{equation}

3637: 

3638: 

3639: The values of $\varepsilon$ and $k$ in~$(\ref{proj1})$ are uniquely

3640: determined by $q'(x)$.

3641: \end{lemma}

3642: 

3643: {\sc Proof.}

3644: 

3645: By the definition of a projective element, $x$ is equivalent to a

3646: diagonal reduced projective element

3647: $$

3648: x_1=\mel{\alpha_1} {\beta_1}{\ A_1} 0, \qquad

3649: {\rm where}\ \alpha_1=\gcd(x_1)=\gcd(x).

3650: $$

3651: under the action of transformations (\ref{elem-tr1}). Since any projective

3652: element is primitive, we have $\alpha_1=1$. In addition, acting by $T(s)$

3653: if necessary we can assume that $A_1$ is in the Smith normal form

3654: (Theorem~$\ref{np-orbits}$).

3655: Summarizing these remarks we conclude that $x_1$ is of the form

3656: \begin{equation}\label{x1}

3657: x_1=\mel 1{\beta_1}{\ \diag{a_1}{a_2}{a_3}}0

3658: \end{equation}

3659: with $\diag{a_1}{a_2}{a_3}$ in the Smith normal form.

3660: 

3661: The definition of the projective element

3662: (Definition~$\ref{def:proja}$(a), see also~$(\ref{def:projexpr})$) implies that

3663: \begin{equation} \label{proj3}

3664: \gcd\{ \beta_1,\ a_1 a_2, \ a_1 a_3, \ a_2 a_3\}=1.

3665: \end{equation}

3666: Since $A_1$ is in the Smith normal form with $a_1 | a_2,\ a_2|a_3$,

3667: the relation (\ref{proj3}) is equivalent to

3668: \begin{equation} \label{proj4}

3669: \gcd\{ \beta_1,\ a_1 a_2\}=1.

3670: \end{equation}

3671: 

3672: 

3673: {\bf Step 1.}

3674: We show that $a_1$ can be taken to be equal to $1$.

3675: 

3676: Assume this is not the case. Then $a_1>1$

3677: and $a_1|a_2, a_1|a_3$ (case $a_1=a_2=a_3=0$ needs slightly

3678: different treatment; we skip details).

3679: Relation (\ref{proj4}) implies $\gcd(\beta_1, a_1)=1$.

3680: 

3681: Then we apply Lemma~\ref{lcomp} (iii) to $x_1$ (with $c=1$)

3682: and transform it to

3683: \begin{equation}\label{proj5}

3684: \mel 1

3685: {\beta_1-2a_1a_2}{\quad \diag{a_1}{\ a_2}{\ a_3+\beta_1-{a_1a_2}}}0,

3686: \end{equation}

3687: It follows from the above that

3688: $\gcd\{{a_1},{a_2},\ a_3+\beta_1-{a_1a_2}\}=

3689: \gcd\{{a_1},{a_2}, \beta_1\}=1$.

3690: The $\gcd$ condition implies that the Smith canonical form

3691: of $$\diag{a_1}{a_2}{\ a_3+\beta_1-{a_1a_2}}$$ looks like

3692: $\diag{1}**$.

3693: Then we can apply an appropriate $T(s)$

3694: to (\ref{proj5}),

3695: and it will yield the desired result, completing the first step.

3696: 

3697: \medskip

3698: Step 1 implies that we can assume that $a_1$ in the element~$(\ref{x1})$

3699: is equal to $1$ and we proceed to

3700: 

3701: {\bf Step 2.}

3702: We show that $a_2$ in~$(\ref{x1})$ can be taken to be equal to $1$.

3703: 

3704: Assume this is not the case. Then $a_2>1$ and we still have $a_2|a_3$

3705: (case $a_2=a_3=0$ treated similarly).

3706: Relation (\ref{proj3}) implies $\gcd(\beta_1, a_2)=1$.

3707: 

3708: We again apply Lemma~\ref{lcomp} (iii) to $x_1$ (with $c=1$)

3709: and transform it to

3710: \begin{equation}\label{proj6}

3711: \mel 1

3712: {\beta_1-2a_2}{\quad \diag{1}{\ a_2}{\ a_3+\beta_1-{a_2}}} 0.

3713: \end{equation}

3714: 

3715: We have $\gcd\{ {a_2}, a_3+\beta_1-{a_2} \}=

3716: \gcd\{ {a_2},\beta_1\}=1$.

3717: Similarly to Step~1, we can apply an appropriate $T(s)$

3718: to (\ref{proj6}), an get an element of the form

3719: $$

3720: \mel 1{\beta_1-2a_2}{\quad \diag{1}{1}{*}}0.

3721: $$

3722: The second step is complete.

3723: 

3724: It follows from the above steps that element (\ref{x1}) may be taken to be

3725: of the form

3726: \begin{equation}\label{proj7}

3727: \mel 1 {\beta_1}{\quad \diag{1}{1}{a_3}}0.

3728: \end{equation}

3729: 

3730: {\bf Step 3.}

3731: We show that $\beta_1$ in the element (\ref{proj7}) may be taken to be $0$ or

3732: $1$.

3733: 

3734: To do it we again apply Lemma~\ref{lcomp} (iii):

3735: $$

3736: \mel 1{\beta_1-2c}{\quad \diag{1}{\ 1}{\ a_3+\beta_1 c-c^2}}0.

3737: $$

3738: 

3739: Obviously, there is a $c$, which makes the second component $0$ or

3740: $1$.

3741: \bigskip

3742: 

3743: The three steps above show that an arbitrary projective element can be

3744: brought to the form~(\ref{proj1}). It remains to prove the uniqueness

3745: assertion.

3746: 

3747: When $x'$ is of the form (\ref{proj1}), we obtain from

3748: $(\ref{qmod})$ that

3749: $$

3750: q'(x') = 4 k+\varepsilon ^2 .

3751: $$

3752: 

3753: This equation determines the parity of $\varepsilon$ uniquely,

3754: and hence $k$ is unique as well.

3755: \ep

3756: 

3757: Now we have all the necessary tools to prove the main result of the paper:

3758: the theorem concerning orbits in the spaces associated to cubic Jordan

3759: algebras of Hermitian matrices over split composition algebras.

3760: 

3761: \begin{theorem}\label{thm:main}

3762: 

3763: Let $(G_\zet, \mz)$ be one the following pairs

3764: $$

3765: \Bigl(\SL_6(\zet), \wedge^3(\zet^6)\Bigr),\quad

3766: \Bigl(D_6(\zet), \mbox{\rm half-spin}_\zet\Bigr),\quad

3767: \Bigl(E_7(\zet), V(\omega_7)_\zet\Bigr).

3768: $$

3769: Then

3770: \begin{itemize}

3771: \item

3772: The $G_\zet$-invariant quartic form (the norm) on the module $\mz$

3773: has values congruent to $0$ or $1\, (\!\mod 4)$.

3774: 

3775: \item

3776: Let $n$ be an integer $\equiv 0$ or $1\, (\!\mod 4)$.

3777: The group $G_\zet$ acts

3778: transitively on the set of {\em projective} elements of norm $n$.

3779: 

3780: \item

3781: If $n$ is a fundamental discriminant\footnote{

3782: An integer $n$ is called a fundamental discriminant if $n$ is squarefree

3783: and $\equiv 1 (\mod 4)$

3784: or $n=4k$, where $k$ is a squarefree integer that is $\equiv 2$ or $3 (\mod 4)$.

3785: The result stated in the theorem also applies in the case $n=1$.

3786: },

3787: then every element of norm $n$ is projective, and hence in this case

3788: $G_\zet$ acts transitively on the set of elements of norm $n$.

3789: \end{itemize}

3790: \end{theorem}

3791: 

3792: {\sc Proof.}

3793: It was shown in Proposition~$\ref{prop:gener}$ that for $\J=\h\bin,

3794: \h\quat, \h\oct$ the Freudenthal construction yields an

3795: absolutely almost simple connected algebraic

3796: group $G=\gr$ of type $A_5, D_6, E_7$, respectively.

3797: Each of these groups acts on the vector space $\m=\mj$, producing

3798: an irreducible representation

3799: whose type is given in the statement of the theorem.

3800: 

3801: The integral structure in the group $G_\zet=\grz$

3802: and the module $\mz$ is induced by the integral structure in $\jz$.

3803: The case $\jz=\h\binz$ requires special treatment.

3804: It was shown in Example~$\ref{ex:sl6int}$ that $\SL_6(\zet)$ is isomorphic

3805: to a subgroup (of index two) in $\g{\m(\h{\bin_\zet})}$. We will do the proof

3806: of the theorem for the whole group $\grz$, and address the issue of

3807: $\SL_6(\zet)$ at the end of the proof.

3808: 

3809: The module $\mz=\m(\jz)$ comes equipped with the quartic form

3810: $$

3811: q'(x)= \Bigl((A,B)-\alpha\beta\Bigr)^2

3812: -4 (A^\#, B^\#) +4\alpha \mnorm(A) +4\beta \mnorm(B),\qquad

3813: x=\mel{\alpha}\beta A B \in \mz,

3814: $$

3815: invariant with respect to $G_\zet$. It was noted in Subsection~$\ref{ss:33int}$

3816: that the operations $ \mnorm$ and $(\cdot,\cdot)$ in $\jz$ have integer

3817: values, which implies that $q'(x)$ is always congruent

3818: to $0$ or $1$ modulo $4$.

3819: 

3820: Next, let $x$ be a projective element in $\mz$. It follows from

3821: Lemma~$\ref{red2}$, that $x$ can be transformed to an element of the

3822: form

3823: \begin{equation}

3824: \mel{1}\varepsilon {\diag{1}1k}0,\qquad

3825: \varepsilon\in\{0,1\},\quad

3826: k\in \zet

3827: \end{equation}

3828: with values of $\varepsilon$ and $k$ uniquely determined by the value of

3829: $q'(x)$. This immediately implies that the group $\grz$ acts transitively

3830: on the set of projective elements of norm $q'(x)$.

3831: 

3832: Finally, let an integer $n$ be a fundamental discriminant, and let

3833: $x\in\mz$ be such that $q'(x)=n$.

3834: By Lemma~$\ref{bl-fz}$, $x$ is equivalent to a diagonal reduced element of the

3835: form

3836: \begin{equation}\label{red-thm}

3837: x_1=\mel \alpha\beta {\diag{a_1}{a_2}{a_3}} 0,\qquad

3838: \alpha>0,\alpha|\beta,\alpha|a_i.

3839: \end{equation}

3840: We need to show that $x$ is projective, and by

3841: Definition~$\ref{def:proja}$(b), it is sufficient to show that $x_1$ is

3842: projective.

3843: 

3844: We have $q'(x)=q'(x_1)$, and hence

3845: \begin{equation}\label{n1}

3846: n=\alpha^2\beta^2+4a_1a_2a_3.

3847: \end{equation}

3848: 

3849: First we show that $\alpha$ must be equal to $1$.

3850: 

3851: Note that $(\ref{red-thm})$ and $(\ref{n1})$ imply $\alpha^3|n$.

3852: 

3853: If $\alpha >2$, this remark implies that $n$ is not square-free. If

3854: $\alpha=2$, we get $16|n$. And in both cases we get that $n$ is not a

3855: fundamental discriminant. Hence $\alpha=1$.

3856: 

3857: It means we can rewrite $x_1$ in the form

3858: $x_1=\mel 1 \beta {\diag{a_1}{a_2}{a_3}} 0$, and

3859: $$

3860: n=\beta^2+4a_1a_2a_3.

3861: $$

3862: 

3863: To complete the proof we will show that if $x_1$ is not projective, then

3864: $n$ is not a fundamental discriminant.

3865: 

3866: So suppose that $x_1$ is not projective. Then the $\gcd$ in one of the

3867: expressions~$(\ref{def:projexpr})$ is greater than $1$. Without loss of

3868: generality (and using $\alpha=1$) we will assume that

3869: $$

3870: \gcd (a_3, \beta, a_1 a_2)>1.

3871: $$

3872: Let $p$ be a prime dividing $\gcd (a_3, \beta, a_1 a_2)$. It follows that

3873: $$

3874: p|\beta \quad \mbox{ and \ $p$ divides at least two of the $a_i$'s}.

3875: $$

3876: This remark implies $p^2|n$. If $p>2$ it already implies that $n$ is not a

3877: fundamental discriminant. And if $p=2$, it implies that

3878: $$

3879: \beta=2\beta_1, \qquad

3880: a_1a_2a_2=4c

3881: $$

3882: for some integers $\beta_1$ and $c$. Hence $n$ may be rewritten in the

3883: form

3884: $$

3885: n=4\Bigl(\beta_1^2+4c\Bigr).

3886: $$

3887: Depending on the parity of $\beta_1$, the quantity $\beta_1^2+4c$ is

3888: congruent to $0$ or $1$ modulo $4$. In both cases it implies that $n$ is

3889: not a fundamental discriminant.

3890: 

3891: We thus proved that if $n$ is a fundamental discriminant, then every

3892: element of norm $n$ is projective, and hence $G_\zet$ acts transitively on

3893: the set of such elements.

3894: \medskip

3895: 

3896: Our last remark is concerned with the case $\jz=\h\binz$. It was noted

3897: in Example~$\ref{ex:sl6int}$ that $\SL_6(\zet)\cong \g{\jz}\cc$, which is

3898: a subgroup of index two in $\grz$. We need to show that orbits under

3899: $\g{\jz}\cc$ are the same as the orbits under the action of the whole

3900: group $G_\zet=\grz$.

3901: 

3902: For this we take an arbitrary element $x\in\mz$ and let $x_1$ be a

3903: diagonal reduced element

3904: $$

3905: x_1=\mel \alpha\beta {\diag{a_1}{a_2}{a_3}} 0,\qquad

3906: \alpha>0,\alpha|\beta,\alpha|a_i

3907: $$

3908: lying in the $\grz$-orbit of $x$, i.e., $x_1=g(x)$ for some $g\in\grz$.

3909: It was noted in Example~$\ref{ex:sl6int}$ that every $g\in\grz$ has the

3910: form

3911: $$

3912: g=g_0\quad

3913: \mbox{or}\quad

3914: g=T(t) g_0

3915: $$

3916: with $g_0\in\grz\cc$. There is nothing to prove in first case, and in the

3917: second case we have

3918: $$

3919: x_1=T(t)(x_1)=T(t)^2\,g_0 (x)=g_0(x)

3920: $$

3921: using that $T(t)$ acts as the transpose operation

3922: on the matrices at the two off-diagonal

3923: entries of $x_1$, and the fact that $T(t)^2=\id_\mz$. Hence $\grz$-orbits

3924: are the same as $\grz\cc$-orbits when $\jz=\h\binz$, and in this case

3925: the theorem remains true for the subgroup of index two isomorphic to

3926: $\SL_6(\zet)$.

3927: \ep

3928: 

3929: 

3930: \subsection{Invariant factors

3931: in the module $\mz$ and the degenerate orbits}

3932: \label{ss:if}

3933: 

3934: In this subsection we introduce the ``invariants" $d_i, i=1,2,3,4$,

3935: reminiscent of invariant factors

3936: for regular matrices over integers, and use them to

3937: describe the orbits of degenerate elements, i.e., those whose norm is

3938: equal to zero.

3939: We define these invariants using the concept of the rank polynomials introduced

3940: earlier in $(\ref{p4})-(\ref{p2})$.

3941: 

3942: \medskip

3943: 

3944: \begin{defn} \label{def:di}

3945: {\rm Let $\jz$ be as in~$(\ref{zlist})$, and let $\mz=\m(\jz)$.

3946: For an element $\bx\in\mz$ of the form

3947: $$

3948: \bx = \mel\alpha\beta A B

3949: $$

3950: the functions $d_i:\ \mz\ \to \ \zet$ are defined by the following

3951: expressions

3952: \begin{itemize}

3953: \item

3954: $d_1 (\bx) =\gcd(\bx)$;

3955: \item

3956: $d_2 (\bx) = \gcd \Bigl(3\,T(x,x,y)+\{x,y\}\,x\Bigr) \quad

3957: \mbox{for all }y\in \mz$;

3958: %

3959: \item

3960: $d_3 (\bx) = \gcd ( T(\bx,\bx,\bx) )$;

3961: \item

3962: $d_4 (\bx) = q'(\bx)$.

3963: \end{itemize}

3964: }

3965: \end{defn}

3966: 

3967: \begin{lemma} \label{di}

3968: Let $\mz$ be as in Definition~$\ref{def:di}$.

3969: \begin{enumerate}[{\rm (i)}]

3970: \item

3971: The functions $d_i$ are invariant under the action of the

3972: group $\grz$.

3973: \item

3974: \begin{equation}\label{d2}

3975: d_2(x)= \gcd\Bigl\{

3976: 3\alpha\beta-(A,B),\quad

3977: 2(\alpha A-B^\# ), \quad 2(\beta B - A^\# ),\quad

3978: 2Q(x)\Bigr\}.

3979: \end{equation}

3980: \end{enumerate}

3981: 

3982: \end{lemma}

3983: 

3984: {\sc Proof.}

3985: 

3986: {\bf (i)}

3987: The statement for $d_1$ follows from Lemma~$\ref{gcd}$.

3988: 

3989: The function $d_4$ is the norm $q'$, and the statement follows from the

3990: definition of the groups $\gr$ and $\grz$.

3991: 

3992: The statement for $d_3$ follows from the following computation

3993: $$

3994: d_3(\eta(x))= d_1\biggl(\  T\Bigl(\eta(x), \eta(x),\eta(x)\Bigr)\

3995: \biggr)=

3996: $$

3997: $$

3998: =d_1\biggl(\ \eta \Bigl( T(x,x,x) \Bigr)\ \biggr)= d_1\biggl( T(x,x,x)

3999: \biggr)= d_3(x) \qquad \mbox{for any }\eta\in \grz.

4000: $$

4001: 

4002: In this computation we used the definition of $d_3$, the relation

4003: $(\ref{autot})$, and the invariance statement for $d_1$.

4004: 

4005: \medskip

4006: The statement for $d_2$ follows from the invariance of the skew-symmetric

4007: form $\{\cdot, \cdot\}$ and the argument analogous to that in the previous

4008: paragraph.

4009: 

4010: 

4011: {\bf (ii)}

4012: The computation ~$(\ref{r1eq})$ of Lemma~$\ref{lem:rank1}$ implies that

4013: $$

4014: \mbox{an integer $d$ divides}\quad

4015: 3\,T(x,x,y)+\{x,y\}\,x \quad

4016: \mbox{for any }y\in \mz

4017: $$

4018: if and only if $d$ divides each of the following expressions

4019: $$

4020: 3\alpha\beta-(A,B),\quad

4021: 2(\alpha A-B^\# ), \quad 2(\beta B - A^\# ),\quad

4022: 2Q(x),\quad 2Q(x'),

4023: $$

4024: where $x=\mel \alpha\beta A B$, $x'=\mel \beta \alpha B A$, and

4025: $Q(x)\in \End_\zet(\jz)$ was defined in~$(\ref{q})$.

4026: 

4027: As in the proof of Lemma~$\ref{lem:rank1}$ we have

4028: $$

4029: Q(x')(C)=Q(x)(C)-2Q(x)(\one)\jprod C\quad

4030: \mbox{for any }C\in \jz.

4031: $$

4032: This implies that $d$ divides $2Q(x')$ whenever $d$ divides $2Q(x)$,

4033: and the statement (ii) of the lemma follows.

4034: \ep

4035: 

4036: 

4037: \begin{rem}\label{rem:rpz}

4038: {\rm

4039: It is possible to define alternative quadratic invariant $\dvap$

4040: via

4041: \begin{equation}\label{alt-d2}

4042: \dvap(x)=\gcd\Bigl(

4043: \alpha A-B^\# , \quad \beta B - A^\# ,\quad Q(\bx)

4044: \Bigr).

4045: \end{equation}

4046: 

4047: However the short proof of $\grz$-invariance of $d_2$ in the previous

4048: lemma does not work for $\dvap$. One can still prove that $\dvap$ is

4049: invariant under each of the transformations $\phi(C), \psi(D), T(s)$, and

4050: then using Proposition~$\ref{genz}$ conclude that $\dvap$ is invariant

4051: under the whole $\grz$. This route is quite technical, it was implemented

4052: in Lemma~$2.3.5$ of \cite{kphd}.

4053: 

4054: Our primary use for the invariants $d_i$ is to distinguish elements of $\mz$

4055: lying in different orbits.

4056: We should note that $\dvap$ is ``finer" than $d_2$ in this sense.

4057: For example, using $\dvap$ one can conclude that elements

4058: $$

4059: \mel 1 2 {\diag{1}00} 0

4060: \quad {\rm and}\quad \mel 1 2 {\diag{2}00} 0

4061: $$

4062: lie in distinct $\grz$-orbits, though all $d_i$'s are equal for these two

4063: elements.

4064: 

4065: In this paper will make use of more ``coarse" relations~$(\ref{d2})$,

4066: which are sufficient for our purposes. \ee

4067: }

4068: \end{rem}

4069: 

4070: \begin{rem} \label{rem2}

4071: {\rm

4072: By analogy with Remark~$\ref{rem1}$ we notice that when $x$ is a reduced

4073: element

4074: $$

4075: x=\mel \alpha\beta A 0

4076: $$

4077: then

4078: $%

4079: d_2(x)=\gcd\Bigl(\alpha\beta,\ 2\alpha A,\ 2A^\# \Bigr)

4080: $. \ee%

4081: }

4082: \end{rem}

4083: 

4084: The following theorem in an extended version of Theorem~$\ref{thm:main}$.

4085: It provides the description of degenerate $\grz$-orbits (corresponding the

4086: zero value of the quartic form) in the module $\mz$.

4087: 

4088: \begin{theorem} \label{thm2}

4089: Let $\jz$ be one of $\h{\binz}, \h{\quatz}, \h{\octz}$

4090: and $\mz=\m(\jz)$.

4091: 

4092: 

4093: \begin{itemize}

4094: 

4095: \item

4096: Every element $\bx$ of rank $1$ in the module $\mz$ can be brought to the

4097: form

4098: \begin{equation} \label{thr1}

4099: \mel \alpha 000 \quad {\rm where}\  \alpha = d_1(\bx)

4100: \end{equation}

4101: by an element in the group $\grz$.

4102: 

4103: The set

4104: \begin{equation} \label{thset1}

4105: \left.\left\{\mel k 0 0 0\ \right|\ k\in\zet, k\ge 1\right\}

4106: \end{equation}

4107: is a complete set of {\em distinct\/} orbit representatives of elements of

4108: rank $1$ in $\mz$.

4109: 

4110: \item

4111: Every element $\bx$ of rank $2$ in the module $\mz$ can be brought to the

4112: form

4113: \begin{equation} \label{thr2}

4114: \mel \alpha 0 {\ \diag{a}00} 0

4115: \quad {\rm where}\  \alpha|a,\ \alpha =

4116: d_1(\bx), \ a= d_2(\bx)/\alpha

4117: \end{equation}

4118: by an element in the group $\grz$.

4119: 

4120: The set

4121: \begin{equation} \label{thset2}

4122: \left\{

4123: \left.\mel k 0 {\diag{m}00} 0 \

4124:  \right.|\ k,m\in\zet, k,m>0, k|m\right\}

4125: \end{equation}

4126: is a complete set of {\em distinct\/} orbit representatives of elements of

4127: rank $2$ in $\mz$.

4128: 

4129: \item

4130: In the case $\rank x=3$ or $4$, the group $\grz$ acts transitively on the

4131: set of {\em projective} elements of norm $n$.

4132: Every such element may brought to the form

4133: \begin{equation}\label{hirank}

4134: \mel 1 \varepsilon{\ \diag{1}1k} 0,\qquad

4135: \varepsilon\in\{0,1\},\quad

4136: k\in \zet,

4137: \end{equation}

4138: where $k=\frac{q'(x)-\varepsilon^2}{4}$ and $\varepsilon\equiv

4139: q'(x)\,(\mod 2)$ are uniquely determined by $x$.

4140: \end{itemize}

4141: \end{theorem}

4142: 

4143: {\sc Proof.}

4144: 

4145: We will often use in the proof the facts that transformations in the group

4146: $\grz$ preserve the rank and the invariants $d_i$ (Lemma~$\ref{lem:rank}$,

4147: $\ref{di}$) of elements of $\mz$.

4148: 

4149: We note that if $\rank x \le 2$, then $T(x,x,x)=0$

4150: (Definition~$\ref{def:rank}$).

4151: Such an $x$ is not projective by Proposition~$\ref{prop:proj}$(ii),

4152: and hence Theorem~$\ref{thm:main}$ gives no information about orbits

4153: of such elements.

4154: 

4155: \medskip

4156: {\em The case of\/} rank $1$.

4157: 

4158: Let $\bx$ be an arbitrary element of rank $1$. First we show that $\bx$

4159: can be brought to the form $(\ref{thr1})$. Then we show that elements of

4160: the form $(\ref{thset1})$ with distinct $k$'s lie in distinct orbits of

4161: the group $\grz$.

4162: 

4163: By Lemma~$\ref{bl-fz}$ there exists $\sigma\in \grz$ such that

4164: $$

4165: \sigma (\bx) = \mel \alpha \beta A 0 ,

4166: \quad \qquad \alpha>0,\

4167: \alpha|\beta,\ \alpha |A.

4168: $$

4169: 

4170: Since rank $\bx =1$, it follows from $(\ref{r1s})$ that $\beta =0$ and

4171: $A=0$. So

4172: $$

4173: \sigma (\bx) = \mel \alpha 000

4174: $$

4175: 

4176: is in the desired form.

4177: 

4178: \medskip

4179: Now suppose we have two elements $\bx_1=\mel {k_1}000$ and

4180: $\bx_2=\mel{k_2}000$

4181: of the form $(\ref{thset1})$ lying in the same $\grz$-orbit. Then

4182: $d_1(\bx _1)=k_1$ and $d_1(\bx _2)=k_2$, and since $d_1$ is constant on

4183: orbits, we have $k_1=k_2$, and hence $x_1=x_2$.

4184: The proof in the case of rank $1$ is complete.

4185: 

4186: 

4187: 

4188: \medskip

4189: {\em The case of\/} rank $2$.

4190: 

4191: Let $\bx$ be an arbitrary element of rank $2$. First we show that $\bx$

4192: can be brought to the form~$(\ref{thr2})$. Then we show that elements of

4193: the form~$(\ref{thset2})$ with distinct $k$'s and $m$'s lie in distinct

4194: orbits of the group $\grz$.

4195: 

4196: By Lemma~$\ref{bl-fz}$ there exists $\sigma\in \grz$ such that

4197: $$

4198: \sigma (\bx) = \mel\alpha\beta A 0, \quad \qquad \alpha>0,\

4199: \alpha|\beta,\ \alpha |A.

4200: $$

4201: 

4202: Since rank $\bx =2$, it follows from $(\ref{r2s})$ that $\beta =0$ and

4203: $A^\# =0$. The last relation implies that rank $A\le 1$. On the other

4204: hand, $A\ne 0$ as $A=0$ would imply that rank $x \le 1$ (see

4205: $(\ref{r1s})$). Hence rank $A=1$.

4206: 

4207: By Theorem~$\ref{np-orbits}$,

4208: there exists $\xi\in \npg\jz$ which brings $A$ to the

4209: Smith normal form. Since $\rank A =1$,

4210: this normal form has only one nonzero (in fact, positive)

4211: element on the diagonal. We thus get

4212: 

4213: $$

4214: \tf{\xi}\, \sigma \ (\bx)= \mel \alpha 0 {\diag{a_1}00} 0.

4215: $$

4216: 

4217: By construction $a_1 = \gcd(A)$, and so $\alpha | a_1$.

4218: 

4219: Hence we have brought $\bx$ to the desired form~$(\ref{thr2})$.

4220: \medskip

4221: 

4222: Now let us take two elements $x_1, x_2$ of the form $(\ref{thset2})$

4223: $$

4224: x_i= \mel {k_i} 0 {\diag{m_i}00} 0\ \qquad k_i,m_i\in\zet, k_i, m_i > 0,

4225: k_i|m_i, i=1,2

4226: $$

4227: 

4228: lying in the same $\grz$-orbit. We want to show that $x_1=x_2$.

4229: 

4230: We have $d_1(x_i)=k_i$, and using the fact that $d_1$ is constant on

4231: orbits, we get

4232: $$

4233: k_1=d_1(x_1)=d_1(x_2)=k_2.

4234: $$

4235: 

4236: Next we notice that Remark~$\ref{rem2}$ implies

4237: $$

4238: d_2(x_i)\ =\ 2\, k_i\, m_i

4239: $$

4240: 

4241: and hence

4242: $$

4243: 2\,k_1\, m_1=d_2(x_1)=d_2(x_2)=2\, k_2\, m_2,

4244: $$

4245: 

4246: which implies that in this case $m_1=m_2$, and hence $x_1=x_2$.

4247: This completes the proof in the case of rank $2$.

4248: 

4249: \medskip

4250: {\em The case of\/} rank $>2$.

4251: A complete reduction procedure is not known in the case of elements of

4252: rank $3$ and $4$, and the structure of $\grz$-orbits may be quite complicated.

4253: For example, when $\jz=\h\binz$, the structure of orbits

4254: in $\mz$ is essentially equivalent to the structure of $\SL_6(\zet)$-orbits

4255: in $\wedge^3(\zet^6)$ (see Example~$\ref{ex:sl6int}$).

4256: It is as complicated as the structure of

4257: (balanced) triples of ideal classes in quadratic rings

4258: \cite[Theorem~$18$]{bh1}.

4259: 

4260: Orbits of the projective elements are the most interesting from the

4261: viewpoint of number theory in this case, and the transitivity result

4262: stated in the theorem follows from Lemma~$\ref{red2}$ (it was also treated

4263: in more detail in Theorem~$\ref{thm:main}$). The case $k=0$ corresponds to

4264: the projective elements of rank~$3$ ($n=q'(x)=0$), and the case $k\ne 0$

4265: corresponds to non-degenerate orbits ($q'(x)\ne 0$).

4266: 

4267: We notice that in the case $\jz=\h\binz$ the result of the theorem remains

4268: valid for the subgroup $\grz\cc$ isomorphic to $\SL_6(\zet)$.

4269: \ep

4270: 

4271: \input{appendix.tex}

4272: 

4273: %

4274: 

4275: 

4276: \begin{thebibliography}{99}

4277: 

4278: \bibitem{b1} %

4279: N. Bourbaki. {\it \'El\'ements de math\'ematique. Groupes et alg\`ebres de Lie.} Chapitre

4280: IV--VI. Hermann, Paris, 1968.

4281: 

4282: \bibitem{bh-phd} %

4283: M. Bhargava. {\em Higher composition laws.} Ph.D. Thesis, Princeton

4284: University, 2001.

4285: 

4286: \bibitem{bhau} %

4287: M. Bhargava. {\em Gauss composition and generalizations.}

4288: Algorithmic Number Theory: 5th International Symposium, Sydney, Australia.

4289: Lecture Notes in Computer Science, 2002.

4290: %

4291: 

4292: \bibitem{bh1} %

4293: M. Bhargava.

4294: {\it Higher composition laws I:

4295: a new view on Gauss composition, and quadratic generalizations.}

4296: Ann. of Math.(2), {\bf 159}(2004), no. 1, pp. 217--250.

4297: 

4298: 

4299: \bibitem{b} %

4300: R. Brown. {\it Groups of type $E_{7}$.} J. Reine Angew. Math. {\bf 236}(1969), pp.

4301: 79--102.

4302: 

4303: \bibitem{clerc} %

4304: J.-L. Clerc.

4305: {\em Special prehomogeneous vector spaces associated

4306: to $F\sb 4, E\sb 6, E\sb 7, E\sb 8$ and

4307: simple Jordan algebras of rank $3$.}

4308: J. Algebra,  {\bf 264}(2003), no. 1, pp. 98--128.

4309: 

4310: \bibitem{dg} %

4311: P. Deligne, B.H. Gross.

4312: {\em On the exceptional series, and its descendants.}

4313: C. R. Math. Acad. Sci. Paris, {\bf 335}(2002), no. 11, pp. 877--881.

4314: 

4315: \bibitem{fau} %

4316: J. Faulkner.

4317: {\em A geometry for $E_7$.}

4318: Trans. Amer. Math. Soc. {\bf 167}(1972), pp. 49--58.

4319: 

4320: \bibitem{fer}  %

4321: J.C. Ferrar.

4322: {\em Strictly regular elements in Freudenthal triple systems.}

4323: Trans. Amer. Math. Soc., {\bf 174}(1972), pp. 313--331 (1973).

4324: 

4325: \bibitem{fr14}  %

4326: H. Freudenthal.

4327: {\em Beziehungen der $E_7$ and $E_8$ zur Oktavenebene.} I--IV.

4328: Indagationes Math.

4329: {\bf 16}(1954) pp. 218--230,

4330: {\bf 16}(1954) pp. 363--368,

4331: {\bf 17}(1955) pp. 151--157,

4332: {\bf 17}(1955) pp. 277--285.

4333: 

4334: \bibitem{gar} %

4335: R.S. Garibaldi.

4336: {\em Structurable algebras and groups of type $E\sb 6$ and $E\sb 7$.}

4337: J. Algebra {\bf 236}(2001), no.~2, pp. 651--691.

4338: 

4339: \bibitem{gw} %

4340: B. Gross, N. Wallach.

4341: {\em On quaternionic discrete series representations, and their

4342: continuations.}

4343: J. Reine Angew. Math. {\bf 481} (1996), pp. 73--123.

4344: 

4345: \bibitem{hum} %

4346: J.E.~Humphreys.

4347: {\em Linear algebraic groups.} GTM 21.

4348: Springer-Verlag, 1975.

4349: 

4350: \bibitem{J3} %

4351: N. Jacobson.

4352: {\em Some Groups of transformations defined by Jordan algebras. III.}

4353: {J.~Reine Angew. Math.} {\bf 207}(1961), pp. 61--85.

4354: 

4355: \bibitem{ja} %

4356: N. Jacobson. {\em Structure and representations of Jordan algebras.}

4357: AMS Colloquium Publications, Vol. XXXIX. Providence, R.I., 1968.

4358: 

4359: \bibitem{jba} %

4360: N. Jacobson.

4361: {\em Basic algebra. I. Second edition.}

4362: W. H. Freeman and Company, New York, 1985.

4363: 

4364: \bibitem{kac} %

4365: V. Kac.

4366: {\em Classification of simple $Z$-graded

4367: Lie superalgebras and simple Jordan superalgebras.}

4368: Comm. Algebra. {\bf 5} (1977), no. 13, pp. 1375--1400.

4369: 

4370: \bibitem{koe} %

4371: M. Koecher. {\em Imbedding of Jordan algebras into Lie algebras. I.}

4372: Amer. J. Math, {\bf 89}(1967), pp. 787--816.

4373: 

4374: \bibitem{k} %

4375: S. Krutelevich. {\em On a canonical form of a $3\times 3$ Hermitian matrix over the

4376: ring of integral split octonions.} J. Algebra, {\bf 253}(2002), no. 2,

4377: pp. 276--295.

4378: 

4379: \bibitem{kphd} %

4380: S. Krutelevich. {\em Orbits of exceptional groups and Jordan systems.}

4381: Ph.D. thesis, Yale University, 2003.

4382: 

4383: \bibitem{lm} %

4384: J. Landsberg, M. Manivel.

4385: {\em The projective geometry of Freudenthal's

4386: magic square.}

4387: J. Algebra, {\bf 239} (2001), no. 2, pp. 477--512.

4388: 

4389: \bibitem{mms} %

4390: J. Maldacena, G. Moore, A. Strominger.

4391: {\it Counting BPS Blackholes in Toroidal

4392: Type II String Theory.}

4393: arXiv:hep-th/9903163.

4394: 

4395: 

4396: 

4397: \bibitem{mc-sp} %

4398: K. McCrimmon.

4399: {\em The Freudenthal-Springer-Tits constructions of exceptional Jordan

4400: algebras.}

4401: Trans. Amer. Math. Soc. {\bf 139} 1969, pp. 495--510.

4402: 

4403: 

4404: \bibitem{atoj} %

4405: K. McCrimmon.

4406: {\em A taste of Jordan algebras.}

4407: Universitext. Springer-Verlag, New York, 2004.

4408: 

4409: \bibitem{newman} %

4410: M. Newman. {\em Integral matrices.}

4411: Pure and Applied Mathematics, v. 45.

4412: Academic Press, 1972.

4413: 

4414: \bibitem{SK} %

4415: M. Sato and T. Kimura.

4416: {\em A classification of irreducible prehomogeneous vector spaces and their

4417: relative invariants.}

4418: Nagoya Math. J. {\bf 65} (1977), pp. 1--155.

4419: 

4420: \bibitem{spr} %

4421: T.A. Springer.

4422: {\em  Characterization of a class of cubic forms.}

4423: Nederl. Akad. Wetensch. Proc. Ser. A 65

4424: = Indag. Math. {\bf 24} 1962, pp. 259--265.

4425: 

4426: \bibitem{sp-alg} %

4427: T.A. Springer.

4428: {\em Jordan algebras and algebraic groups.}

4429: Springer-Verlag, 1973 (reprinted in 1998).

4430: 

4431: \bibitem{SpV} %

4432: T.A. Springer, F.D. Veldkamp.

4433: {\em Octonions, Jordan algebras and exceptional groups.}

4434: Springer-Verlag, Berlin, 2000.

4435: 

4436: \bibitem{tits} %

4437: J. Tits.

4438: {\em Une classe d'alg\`ebres de Lie en relation avec les alg\`ebres de

4439: Jordan.}

4440: Nederl. Akad. Wetensch. Proc. Ser. A 65 = Indag. Math.,

4441: {\bf 24} 1962, pp. 530--535.

4442: 

4443: \end{thebibliography}

4444: 

4445: {\sc

4446: Department of Mathematics and Statistics,

4447: University of Ottawa

4448: 

4449: 585 King Edward Ave.,

4450: Ottawa, ON, K1N 6N5, Canada

4451: }

4452: 

4453: \medskip

4454: E-mail: {\tt sergei.krutelevich@science.uottawa.ca}

4455: \end{document}

4456: