1: % 01 March 2005. slight revise following the referee'd report. 

2: % 10july04 [post-arXiv] (Theorem 4.1; Appendix B, figure 10 numbering fix)

3: % Part I, 9 jun 04 [rlg changes]

4: % 8 jun 04 [parameter space]

5: % 5 jun 04 [cyan, allan, clm comments]

6: %**************************************************************************

7: % 10 Figure files needed:

8: %    1-std.ps  2-configs.ps

9: %    3-strongly.ps  4-reflection.ps

10: %    5-inversion.ps 6-dual.ps

11: %    7-stagezero.ps 8-stageone.ps

12: %    B1-special.ps  B2-dual.ps

13: %

14: %  Need macros: diagrams.sty 

15: %               diagrams.tex 

16: %  (for package: diagrams)

17: %

18: %

19: %*****************************************************************************

20: % New results (over 2001 version)

21: %      Theorem 3.1 (``only if'' corrected)

22: %      Theorem 4.1, Theorem 4.2

23: %      Theorem 5.1 (no proof); Theorem 5.3

24: %      Appendix B; Prop. 8.1

25: %

26: %

27: %*****************************************************************************

28: \documentclass[11pt]{article}

29: %\usepackage{showlabels}

30: 

31: 

32: \usepackage{amssymb,diagrams}

33: \usepackage{epsfig,amsmath,latexsym}

34: \usepackage{amsfonts}

35: 

36: \setlength{\textwidth}{6.2in}

37: \setlength{\textheight}{8.5in}

38: \setlength{\oddsidemargin}{.2in}

39: \setlength{\topmargin}{-0.25in}

40: \setlength{\headheight}{0in}

41: 

42: \newtheorem{theorem}{Theorem}[section]

43: \newtheorem{cor}[theorem]{Corollary}

44: \newtheorem{defi}{Definition}[section]

45: \newtheorem{lemma}[theorem]{Lemma}

46: \newtheorem{prop}[theorem]{Proposition}

47: \newtheorem{conj}{Conjecture}[section]

48: 

49: 

50: \def\binom#1#2{{#1}\choose{#2}}

51: 

52: \def\slfrac#1#2{\hbox{\kern.1em %

53:  \raise.5ex\hbox{\the\scriptfont0 #1}\kern-.11em %

54:  /\kern-.15em\lower.25ex\hbox{\the\scriptfont0 #2}}}

55: 

56: \newarrow{Into}C--->

57: \newarrow{Line}-----

58: 

59: \newcommand{\ep}{\epsilon}

60: \newcommand{\pf}{\noindent{\bf Proof.~}}

61: \newcommand{\beq}{\begin{eqnarray}}

62: \newcommand{\eeq}{\end{eqnarray}}

63: \newcommand{\beql}[1]{\begin{eqnarray}\label{#1}}

64: \newcommand{\beqs}{\begin{eqnarray*}}

65: \newcommand{\eeqs}{\end{eqnarray*}}

66: \newcommand{\eqn}[1]{(\ref{#1})}

67: 

68: \newcommand{\bdelta}{\mbox{\boldmath $\delta$}}

69: \newcommand{\bj}{\mathbf j}

70: \newcommand{\bone}{\mathbf 1}

71: \newcommand{\lf}{\lfloor}

72: \newcommand{\rf}{\rfloor}

73: \newcommand{\lc}{\lceil}

74: \newcommand{\rc}{\rceil}

75: 

76: \newcommand{\CC}{{\mathbb C}}

77: \newcommand{\DD}{{\mathbb D}}

78: \newcommand{\HH}{{\mathbb H}}

79: \newcommand{\NN}{{\mathbb N}}

80: \newcommand{\PP}{{\mathbb P}}

81: \newcommand{\QQ}{{\mathbb Q}}

82: \newcommand{\RR}{{\mathbb R}}

83: \newcommand{\WW}{{\mathbb W}}

84: \newcommand{\ZZ}{{\mathbb Z}}

85: 

86: 

87: 

88: \newcommand{\cc}{{\mathbb C}}

89: \newcommand{\rr}{{\mathbb R}}

90: \newcommand{\zz}{{\mathbb Z}}

91: \newcommand{\qq}{{\mathbb Q}}

92: \newcommand{\nn}{{\mathbb N}}

93: \newcommand{\dd}{{\mathbb D}}

94: 

95: \newcommand{\fc}{{\mathfrak c}}

96: \newcommand{\fd}{{\mathfrak d}}

97: \newcommand{\fe}{{\mathfrak e}}

98: \newcommand{\fg}{{\mathfrak g}}

99: \newcommand{\fh}{{\mathfrak h}}

100: \newcommand{\fj}{{\mathfrak j}}

101: \newcommand{\fk}{{\mathfrak k}}

102: \newcommand{\fp}{{\mathfrak p}}

103: \newcommand{\fq}{{\mathfrak q}}

104: \newcommand{\fr}{{\mathfrak r}}

105: \newcommand{\fs}{{\mathfrak s}}

106: \newcommand{\ft}{{\mathfrak t}}

107: \newcommand{\fx}{{\mathfrak x}}

108: \newcommand{\fy}{{\mathfrak y}}

109: 

110: \newcommand{\ba}{{\mathbf a}}

111: \newcommand{\baa}{\tilde{\mathbf a}}

112: \newcommand{\bb}{{\mathbf b}}

113: \newcommand{\bc}{{\mathbf c}}

114: \newcommand{\br}{{\mathbf r}}

115: \newcommand{\be}{{\mathbf e}}

116: \newcommand{\bff}{{\mathbf f}}

117: \newcommand{\bg}{{\mathbf g}}

118: \newcommand{\bh}{{\mathbf h}}

119: \newcommand{\bm}{{\mathbf m}}

120: \newcommand{\bo}{{\mathbf 1}}

121: \newcommand{\bs}{{\mathbf s}}

122: \newcommand{\bv}{{\mathbf v}}

123: \newcommand{\bw}{{\mathbf w}}

124: \newcommand{\bx}{{\mathbf x}}

125: \newcommand{\by}{{\mathbf y}}

126: \newcommand{\bz}{{\mathbf z}}  

127: 

128: 

129: \newcommand{\bA}{{\mathbf A}}

130: \newcommand{\bB}{{\mathbf B}}  

131: \newcommand{\bC}{{\mathbf C}}

132: \newcommand{\bD}{{\mathbf D}}

133: \newcommand{\bI}{{\mathbf I}}

134: \newcommand{\bJ}{{\mathbf J}}

135: \newcommand{\bK}{{\mathbf K}}

136: \newcommand{\bM}{{\mathbf M}}

137: \newcommand{\bN}{{N}}

138: \newcommand{\bP}{{\mathbf P}}

139: \newcommand{\bQ}{{\mathbf Q}}

140: \newcommand{\bS}{{\mathbf S}}

141: \newcommand{\bT}{{\mathbf T}}

142: \newcommand{\bU}{{\mathbf U}}

143: \newcommand{\bV}{{\mathbf V}}

144: \newcommand{\bW}{{\mathbf W}}

145: \newcommand{\bX}{{\mathbf X}}

146: \newcommand{\bY}{{\mathbf Y}}

147: \newcommand{\bZ}{{\mathbf Z}}

148: 

149: \newcommand{\asc}{{\bf asc}}

150: 

151: \newcommand{\sA}{{\mathcal A}}

152: \newcommand{\sC}{{\mathcal C}}

153: \newcommand{\sD}{{\mathcal D}}

154: \newcommand{\sE}{{\mathcal E}}

155: \newcommand{\sF}{{\mathcal F}}

156: \newcommand{\sG}{{\mathcal G}}

157: \newcommand{\sI}{{\mathcal I}}

158: \newcommand{\sJ}{{\mathcal J}}

159: \newcommand{\sL}{{\mathcal L}}

160: \newcommand{\sM}{{\mathcal M}}

161: \newcommand{\sP}{{\mathcal P}}

162: \newcommand{\sS}{{\mathcal S}}

163: \newcommand{\sW}{{\mathcal W}}

164: \newcommand{\sX}{{\mathcal X}}

165: \newcommand{\PD}{\sP_{\sD}}

166: \newcommand{\GD}{\sG_{\sD}^0}

167: \newcommand{\bsq}{\vrule height .9ex width .8ex depth -.1ex}

168: 

169: 

170: \makeatletter

171: % put a period after section or subsection number in header

172: \def\@sect#1#2#3#4#5#6[#7]#8{\ifnum #2>\c@secnumdepth

173:      \def\@svsec{}\else

174:      \refstepcounter{#1}\edef\@svsec{\csname the#1\endcsname.\hskip .75em }\fi

175:      \@tempskipa #5\relax

176:       \ifdim \@tempskipa>\z@

177:         \begingroup #6\relax

178:           \@hangfrom{\hskip #3\relax\@svsec}{\interlinepenalty \@M #8\par}%

179:         \endgroup

180:        \csname #1mark\endcsname{#7}\addcontentsline

181:          {toc}{#1}{\ifnum #2>\c@secnumdepth \else

182:                       \protect\numberline{\csname the#1\endcsname}\fi

183:                     #7}\else

184:         \def\@svsechd{#6\hskip #3\@svsec #8\csname #1mark\endcsname

185:                       {#7}\addcontentsline

186:                            {toc}{#1}{\ifnum #2>\c@secnumdepth \else

187:                              \protect\numberline{\csname the#1\endcsname}\fi

188:                        #7}}\fi

189:      \@xsect{#5}}

190: % put a period after theorem and theorem-like numbers

191: \def\@begintheorem#1#2{\it \trivlist \item[\hskip \labelsep{\bf #1\ #2.}]}

192: 

193: \def\plain{plain}\ifx\fmtname\plain\csname fi\endcsname

194:      \def\batchfile{here.doc}

195:      \input docstrip

196:      \preamble

197: 

198:      Do not distribute the stripped version of this file.

199:      The checksum in the header refers to the documented version.

200: 

201:      \endpreamble

202:      \generateFile{here.sty}{t}{\from{here.doc}{}}

203:      \endinput

204: \fi

205: \ifcat a\noexpand @\let\next\relax\else\def\next{%

206:     \documentstyle[here,doc]{article}\MakePercentIgnore}\fi\next

207: \ifx\@Hxfloat\@Hundef\else\expandafter\endinput\fi

208: \let\@Hxfloat\@xfloat

209: \def\@xfloat#1[{\@ifnextchar{H}{\@HHfloat{#1}[}{\@Hxfloat{#1}[}}

210: \def\@HHfloat#1[H]{%

211: \expandafter\let\csname end#1\endcsname\end@Hfloat

212: \vskip\intextsep\vbox\bgroup\def\@captype{#1}\parindent\z@

213: \ignorespaces}

214: \def\end@Hfloat{\egroup\vskip \intextsep}

215: 

216: \makeatother

217: 

218: \renewcommand{\theequation}{\arabic{section}.\arabic{equation}}

219: 

220: \catcode`\@=11

221: %\renewcommand{\section}{

222: %        \setcounter{equation}{0}

223: %        \@startsection {section}{1}{\z@}{-3.5ex plus -1ex minus

224: %        -.2ex}{2.3ex plus .2ex}{\large\bf}

225: %        }

226: 

227: \catcode`@=12

228: 

229: 

230: \thispagestyle{empty}

231: 

232: \begin{document}

233: 

234: 

235: 

236: % TITLE PAGE and ABSTRACT

237: 

238: \begin{center}

239: {\Large {\bf Apollonian Circle Packings: Geometry and Group Theory \\

240: I. The Apollonian Group}}\\

241: 

242: \vspace{1.5\baselineskip}

243: {\em Ronald L. Graham} \\

244: Department of Computer Science and Engineering \\

245:  University of California at San Diego, 

246: La Jolla, CA 92093-0114 \\

247: \vspace*{1.5\baselineskip}

248: 

249: {\em Jeffrey C. Lagarias} \\

250: Department of Mathematics \\

251: University of Michigan, 

252: Ann Arbor, MI 48109--1109 \\

253: \vspace*{1.5\baselineskip}

254: 

255: {\em Colin L. Mallows} \\

256: Avaya Labs, Basking Ridge, NJ 07920 \\

257: \vspace*{1.5\baselineskip}

258: 

259: {\em Allan R. Wilks} \\

260: AT\&T Labs, Florham Park, NJ 07932-0971 \\

261: \vspace*{1.5\baselineskip}

262: 

263: 

264: 

265: {\em Catherine H. Yan}

266: \footnote{

267: Partially supported by NSF grants DMS-0070574, DMS-0245526 and a Sloan

268: Fellowship. This author is also affiliated with Dalian University of

269: Technology, China.}\\

270: Department of Mathematics \\

271: Texas A\&M University, College Station, TX 77843-3368\\

272: \vspace*{1.5\baselineskip}

273: %\today

274: %{\bf (Working Draft: Not for Distribution)} \\

275: \vspace*{1.5\baselineskip}

276: (March 10, 2005  version) \\

277: \vspace*{1.5\baselineskip}

278: {\bf ABSTRACT}

279: \end{center}

280: Apollonian circle packings arise by repeatedly filling the interstices

281: between four mutually tangent circles with further tangent circles.

282: We observe that there exist 

283: Apollonian  packings which have strong integrality

284: properties, in which  all  circles in the packing have

285:  integer curvatures and

286: rational centers such that (curvature)$\times$(center) is an

287: integer vector. This series of papers explain such properties.

288: 

289: A {\em Descartes configuration} is a set of four mutually tangent

290: circles with disjoint interiors. 

291:  An Apollonian circle packing can be described in terms of

292: the Descartes configuration it contains. We describe the 

293:  space of all  ordered, oriented Descartes configurations

294: using  a coordinate system $\sM_\DD$

295: consisting of  those $4 \times 4$

296: real matrices $\bW$ with $\bW^T \bQ_{D} \bW = \bQ_{W}$ where 

297: $\bQ_D$ is the matrix of the Descartes quadratic form

298: $Q_D= x_1^2 + x_2^2+ x_3^2 + x_4^2 - 

299: \frac{1}{2}(x_1 +x_2 +x_3 + x_4)^2$ 

300: and  $\bQ_W$ of the quadratic form

301: $Q_W = -8x_1x_2 + 2x_3^2 + 2x_4^2$. 

302: On the parameter space

303: $\sM_\DD$ the  group

304: $Aut(Q_D)$ acts on the left, and $Aut(Q_W)$ acts on the right,

305: giving two different ``geometric'' actions.

306: Both these  groups are  isomorphic to the 

307: Lorentz group $O(3, 1)$. 

308: The right action of $Aut(Q_W)$ 

309: (essentially) corresponds to Mobius transformations acting on the underlying

310: Euclidean space $\rr^2$ while the left action of $Aut(Q_D)$ is defined

311: only on the parameter space. 

312: 

313: We observe that

314: the Descartes configurations in each Apollonian packing 

315: form  an orbit of a single Descartes

316: configuration under a certain finitely generated discrete

317: subgroup of  $Aut(Q_D)$, 

318: which we call the {\em Apollonian group}.

319: This group consists of 

320: $4 \times 4$ integer matrices, and its integrality properties 

321: lead to the integrality

322: properties observed in some Apollonian circle packings.

323: 

324: We introduce two more related finitely generated groups

325: in   $Aut(Q_D)$, the {\em  dual Apollonian group} produced 

326: from the Apollonian

327: group by a ``duality'' conjugation, and the {\em super-Apollonian group} 

328: which is the group generated by the Apollonian and

329: dual Apollonian groups together. 

330: These groups also

331: consist of integer $4 \times 4$ matrices. We show

332: these groups are hyperbolic Coxeter groups.

333: 

334: 

335: \vspace*{1.5\baselineskip}

336: \noindent

337: Keywords: Circle packings, Apollonian circles, Diophantine equations, Lorentz

338: group, Coxeter group

339: \newpage

340: \setcounter{page}{1}

341: 

342: 

343: %*****************************************************************************

344: %

345: % SECTION 1. 

346: %

347: %*****************************************************************************

348: %

349: \setlength{\baselineskip}{1.0\baselineskip}

350: 

351: \section{Introduction}

352: \setcounter{equation}{0}

353: An Apollonian circle packing is a packing of circles 

354: arising  by repeatedly filling the interstices

355: between four mutually tangent circles with further tangent circles.

356: We call an initial arrangement of four mutually tangent circles with

357: distinct tangents (necessarily $6$ of them) a

358: {\em Descartes configuration}.

359: 

360: %

361: %

362: %  Figure 1

363: %

364: 

365: 

366: \begin{figure}[htbp]

367: \centerline{\epsfxsize=4.0in \epsfbox{1-std.ps}}

368: \caption{An Apollonian packing, labelled with circle curvatures}\label{std}

369: \end{figure}

370: 

371: Starting from any Descartes configuration,

372: we  can recursively construct an  infinite circle packing of the 

373: Euclidean plane, 

374:  in which new circles are added 

375: which are tangent to three of the circles that have already been

376: placed and have interiors disjoint from

377: any of them. The infinite packing obtained in the limit of adding

378: all possible such circles is called an

379: {\em Apollonian packing}. The new circles added

380: at each stage can be

381: obtained using M\"{o}bius transformations of 

382: Descartes configurations in the partial packing. 

383: 

384: An Apollonian packing is pictured in Figure \ref{std}, in

385: which each circle is labelled by its {\em curvature,} which is

386: the inverse of its radius.  The initial

387: configuration consists of two circles of radius $\frac{1}{2}$

388: inscribed in a circle of radius $1$, the latter being assigned

389: negative curvature $-1$ by a convention given in \S3.

390: This particular Apollonian

391: packing has the special property that all circles in the packing

392: have integer curvatures. We call a packing with this

393: property {\em integral.} More remarkably,

394: if one regards the circle centers 

395: as complex numbers, then one can place the initial circles so

396: that every circle in the packing has ``curvature$\times$center''

397: a Gaussian integer (element of the ring $\zz[i].)$ This occurs 

398: for example when the

399: center of the outer circle is placed at $\bz= 0$ and the

400: two circles of radius $\frac{1}{2}$ have centers at $\bz= -\frac{1}{2}$

401: and at $\bz= \frac{1}{2}$. We call a packing

402: with this extra property {\em strongly integral}. 

403: 

404: The object of this paper is to give a  geometric explanation of

405: the origin of these integrality properties involving both

406: the curvatures and the circle centers.

407: This is based on five facts:

408: 

409: (1) An Apollonian packing can be described in terms of

410: the Descartes configurations it contains.

411: 

412: (2) There is a coordinate representation of the set $\sM_{\DD}$

413: of all ordered, oriented Descartes configurations 

414: as a six-dimensional  real-algebraic variety.

415: It consists of  the set of  $4 \times 4$ real matrices $\bW$ satisfying a

416: system of quadratic equations, $\bW^T \bQ_{D} \bW = \bQ_{W}$,

417: which state that $\bW$ conjugates 

418: the Descartes quadratic form $Q_{D}= x_1^2 + x_2^2+ x_3^2 + x_4^2 - 

419: \frac{1}{2}(x_1 +x_2 +x_3 + x_4)^2$

420: to a quadratic  form 

421: $Q_{W}= -8x_1x_2 + 2x_3^2 + 2x_4^2$ that

422: we call the Wilker quadratic form, after Wilker \cite{Wi81}. 

423: This coordinate system we call ``augmented curvature-center

424: coordinates'', 

425: as it encodes the curvatures and centers

426: of the circles in the configuration. We term

427:  $\sM_{\DD}$ with these coordinates the {\em parameter space} of 

428: ordered, oriented Descartes configurations.

429: 

430: (3) The variety $\sM_{\DD}$ is a principal

431: homogeneous space for the Lorentz group 

432: $O(3,1)$, under both a left and a right action of this

433: group, realized as $Aut(Q_{D})$ and $Aut(Q_{W}),$

434: respectively. The right action corresponds to M\"{o}bius

435: transformations acting on the plane, while  the left

436: action acts only on the Descartes configuration space.  

437: 

438: (4) There is a discrete subgroup of the left action

439: by $Aut(Q_{D})$,  the {\em Apollonian group} $\sA$,

440: having the property that the (unordered, unoriented) 

441: Descartes configurations

442: in any Apollonian packing are described by a single

443: orbit of this group. If the Descartes configurations

444: are regarded as ordered and oriented, then 

445: exactly $48$ such orbits correspond to each packing,

446: each one containing a copy of each unordered,

447: unoriented Descartes configuration in the packing.

448: 

449: (5) The Apollonian group $\sA$ consists of  integer

450: $4 \times 4$ matrices.

451: 

452: The last property (5) explains the existence of Apollonian

453: packings having integrality properties of both curvatures

454: and centers. If an initial Descartes configuration has

455: integral augmented curvature-center coordinates, then the 

456: same property persists for all Descartes configurations in the

457: packing, hence for all circles in the packing. The packing

458: pictured in Figure~\ref{std} has this property.

459: 

460: The observation that there are Apollonian packings having

461: all curvatures integral is an old one, and was noted by

462: F. Soddy \cite{Sod36}, \cite{Sod37}, who also gave  an

463: extension to three dimensions, the ``bowl of integers''.

464: The existence of an integral matrix group  explaining such  curvatures

465: was studied in S\"{o}derberg \cite{So92} in 1992. 

466: There are many other mentions in the literature. What we 

467: add in this direction  is the observation that the circle centers

468: can simultaneously have integrality properties. 

469: 

470: A main contribution of these papers is the viewpoint 

471: that these properties should be understood in

472: terms of group actions on the

473: parameter space $\sM_{\DD}$ of all 

474: ordered, oriented Descartes configurations.

475: Besides introducing 

476: coordinate systems for this parameter space

477: (given first in \cite{LMW02} by three of the authors), 

478: we  study the relevant discrete group actions in detail.

479: In particular we introduce a larger group of integer

480: matrices acting on the left, the {\em super-Apollonian group}

481: $\sA^{S}$, whose generators have natural geometric interpretations

482: in terms of their action on Descartes configurations. 

483: We prove that this group is a hyperbolic Coxeter group.

484: 

485: The detailed contents of part I are summarized in

486: the next section. Below we briefly

487: indicate the contents of parts II and III.

488: 

489: In part II we study the integrality properties

490: of the Apollonian packing in more detail, 

491: and their relation to the super-Apollonian group. 

492: We show that every integral Apollonian packing can

493: be transformed to a strongly integral one by a Euclidean

494: motion. We introduce super-packings as orbits of

495: the super-Apollonian group, starting

496: from a given Descartes configuration, and geometric

497: super-packings which consist

498: of  the set of circles in all these

499: Descartes configurations. We show that there 

500: are exactly 8 different primitive

501: \footnote{An integral packing is {\em primitive} if

502: the greatest common divisor of all curvatures of 

503: circles in the packing is $1$.}

504: strongly integral geometric super-packings,

505: and that each of these contains a copy of every

506: primitive integral Apollonian packing. 

507: We  characterize

508: the set of all (primitive) strongly integral Descartes configurations

509: as a collection of $384$ orbits of the super-Apollonian group.

510: 

511: In part III we consider to what extent the results

512: proved in parts I and II extend to higher dimensions.  

513: In all dimensions $n \ge 3$ there are analogues

514: of the parameter space of Descartes configurations, of

515: left and right group actions by $O(n+1, 1)$, and of

516: the Apollonian, dual Apollonian  and super-Apollonian groups. 

517: However in dimensions $n \ge 4$ the Apollonian group

518: action does not correspond to a sphere packing; the

519: spheres will overlap. Nevertheless one can still study the

520: orbits of these group actions on the parameter space.

521: The Apollonian group in higher dimensions has rational

522: entries, rather than integer entries. We show that 

523: configurations having all curvatures rational exist in

524: all dimensions, and having  curvature$\times$centers

525: rational exist in certain dimensions only.

526: 

527: The general framework of these papers was developed

528: by the second author (JCL), who also did much of

529: the writing. This paper is an extensively  revised version of a preprint

530: written in 2000, which adds some new results in \S4 and S5. 

531: 

532: \paragraph {Acknowledgments.}

533: The authors thank Ludwig Balke for suggesting that the 

534: two group actions in \S3 should correspond to a 

535: left action and a right action. The authors are  grateful for

536: helpful comments from Andrew Odlyzko, Eric Rains, 

537: Jim Reeds and Neil Sloane.  The authors thank the reviewer

538: for incisive comments leading to reorganization of the paper.

539: 

540: %*****************************************************************************

541: %

542: % SECTION 2. 

543: %

544: %*****************************************************************************

545: %

546: 

547: \section{Summary of  Results}

548: \setcounter{equation}{0}

549: 

550: 

551: There are at least three  ways to 

552: describe the Apollonian packing $\sP$ containing a given

553: Descartes configuration $\sD$. \\

554: 

555: (G1) [Geometric] An Apollonian  packing $\sP_{\sD}$ 

556: is a set of circles in 

557: $\hat{\cc} = \RR^2 \cup \{ \infty\}$, which comprise four orbits

558: under the action of a discrete group $G_{\sA}(\sD)$ of

559: M\"{o}bius transformations

560: inside  the conformal group M\"{o}b(2). The discrete

561: group $G_{\sA}(\sD)$ depends on $\sD$. \\

562: 

563: (G2) [Algebraic] The collection of all (ordered, oriented) Descartes

564: configurations in the packing $\sP_{\sD}$ form $48$ orbits

565: of a discrete group $\sA$, the Apollonian group, contained

566: in the group $Aut(\sM_{\DD}) \equiv Aut(Q_{D})$ of

567: left-automorphisms of the parameter space $\sM_{\DD}$ of Descartes

568: configurations. The discrete group $\sA$ is independent of 

569: the configuration $\sD$. \\

570: 

571: (G3) [Holographic]   

572: The open disks comprising the interiors of the

573: circles in the packing are the complement

574: $\hat{\cc} \backslash \Lambda_{\sD}$ of the limit

575: set $\Lambda_{\sD}$ of a certain Schottky group $\sS_{\sD}$

576: acting on hyperbolic $3$-space $\HH^3$, with  $\hat{\cc}$

577: identified with  its ideal boundary. 

578: The Schottky group  $\sS_{\sD}$ depends on $\sD$. \\

579: 

580: In this paper we will mainly consider viewpoints

581: (G1) and (G2). Viewpoint (G3) is described in 

582: Mumford, Series and Wright \cite[Chapter 7]{MSW02},

583: and we treat it in  Appendix B. It is termed

584: ``holographic'' because it views the 

585: limit set of the packing as 

586: the boundary of a higher-dimensional object, 

587: which in principle  gives information about it. In

588: Appendix B we point out 

589: a connection of viewpoint (G3) to viewpoint (G1).

590: %which however is not completely understood.

591: The term ``algebraic'' for (G2) refers to the group action

592: being on a real-algebraic variety  of Descartes 

593: configurations. 

594:  

595: 

596: The main emphasis of this series of papers 

597: is to study Apollonian packings in terms of the 

598: Descartes configurations they contain.

599: That is, we  study the packing

600: as a collection of points

601:  inside the parameter space of all 

602: ordered, oriented Descartes configurations. 

603: There are  two different group actions on this parameter space,

604: which are a right action associated to viewpoint (G1)

605: and a left action corresponding to viewpoint (G2), 

606: with the groups both isomorphic to

607: the Lorentz group $O(3,1)$, a real Lie group, as explained

608: in \S3. In particular  group actions are independent and 

609: mutually commute.

610: The discrete group $G_{\sA}(\sD)$ above is contained in 

611: the right action and the discrete group $\sA$, the Apollonian

612: group,  is contained in the  left action.

613: Thus viewpoints (G1) and (G2) are complementary and

614: coexist simultaneously on the space of all Descartes configurations.

615: 

616: 

617: As stated in the introduction,

618: the integer structures in the curvatures

619: and centers of some Apollonian packings can be explained

620: in terms of the viewpoint (G2),

621: using the discrete group $\sA$,

622: the Apollonian group.  This viewpoint can be traced  back to

623: the ``inversive crystal'' in Wilker~\cite[\S 14]{Wi81}.

624: 

625: 

626: In \S3 we coordinatize 

627: the space  $\sM_{\DD}$ of

628: (ordered, oriented) Descartes configurations as in \cite{LMW02}

629: and describe two group actions on this space.

630: In \S3.1 we give two coordinate systems.

631: The first labels such a Descartes configuration with a 

632: $4 \times 3$ matrix $\bM_{\sD}$, 

633: called {\em curvature-center coordinates}, 

634: and 

635: the second with a $4 \times 4$ matrix $\bW_{\sD}$,

636: called {\em augmented curvature-center coordinates}. 

637: These coordinates

638: are characterized  by 

639: quadratic relations which generalize the Descartes circle

640: theorem. Theorem~\ref{th31} exactly characterizes

641: these relations for curvature-center coordinates, 

642: strengthening Theorem 3.2 in \cite{LMW02}

643: by formulating and proving a converse.

644: For the augmented curvature-center coordinates

645: these quadratic relations (proved in  \cite{LMW02})

646: take the form

647: $$

648: \bW_{\sD}^{T} \bQ_{D} \bW_{\sD} = \bQ_{W},

649: $$

650: which gives  a conjugacy  of the Descartes quadratic form $Q_{D}$

651: to the Wilker quadratic form $Q_{W}$.

652: These forms are indefinite of signature $(3,1)$ and their

653: (real) automorphism groups $Aut(Q_{D})$ and $Aut(Q_W)$

654: are isomorphic to the Lorentz group $O(3,1)$.

655: 

656: 

657: 

658: In \S3.2 we describe  a linear  left action (``Lorentz action'')

659: by $Aut(Q_{D})$ and a right action (``M\"{o}bius action'') 

660: by $Aut(Q_W)$ on the space  $\sM_{\DD}$.

661: Theorem~\ref{th33} describes these actions. It shows

662: that the space $\sM_{\DD}$ is a principal homogeoneous

663: space for the group $O(3,1)$ under both the left action

664: and the right action. 

665: The M\"{o}bius action

666: is treated in more detail in Appendix A.

667: In \S3.3  we describe some integral elements of 

668: the Lorentz action $Aut(Q_{D})$ which have geometric

669: interpretations as simple transformations  of a Descartes

670: configuration. These elements  are used in defining the 

671: Apollonian group, dual Apollonian group and super-Apollonian

672: group given below. 

673: 

674: 

675: In \S4 we describe Apollonian packings and the 

676: Apollonian group. Theorem~\ref{th41} establishes

677: the basic fact that the interiors of all circles

678: in an Apollonian packing are disjoint.

679: The much-studied {\em residual set} $\Lambda(\sP)$ of an

680: Apollonian packing is the complement of the

681: interiors of all circles; it is a set of 

682: measure zero.  For later use, 

683: Theorem~\ref{th42} gives several properties  of  the

684: residual set $\Lambda(\sP)$.

685: Theorem~\ref{th41}  shows that the 

686: (ordered, oriented) Descartes configurations

687: in a packing form 48 orbits of this group.

688: 

689: In \S5 we define the dual Apollonian group $\sA^{\perp}$

690: and call its orbits {\em dual Apollonian packings}.

691: Theorem \ref{th43} shows that the  set of all

692: tangency points of circles in a dual Apollonian packing

693: has closure the limit set of another Apollonian packing,

694: that generated by the dual Descartes configuration of

695: any Descartes configuration generating the packing. 

696: 

697: In \S6 we define the super-Apollonian group $\sA^{S}$ to

698: be the group generated by $\sA$ and $\sA^{\perp}$ combined,

699: and call its orbits Apollonian super-packings.

700: It is a discrete group of integer matrices contained in $Aut(Q_D)$.

701: Theorem~\ref{Sth47} in \S6.1

702: gives  a complete  presentation for $\sA^{\perp}$, 

703: establishing that it is a hyperbolic Coxeter group.

704: In \S6.2 we add remarks on  super-packings, which are

705: studied at length in part II.

706: 

707: 

708: 

709: In Appendix A we describe the M\"{o}bius group action in

710: detail. Theorem~\ref{th91} gives an isomorphism of this group to 

711: $Aut(Q_{W})$. 

712: 

713: In Appendix B we describe  the Schottky group

714: action in Mumford, Series and Wright \cite{MSW02}.

715: We indicate some relations to the  

716: M\"{o}bius group action.

717: 

718: %%*****************************************************************************

719: %

720: % SECTION 3. 

721: %

722: %*****************************************************************************

723: %

724: 

725: \section{Descartes Configurations and Group Actions}

726: \setcounter{equation}{0}

727: 

728: 

729: %*****************************************************************************

730: %

731: % SECTION 3.1 

732: %

733: %*****************************************************************************

734: 

735: \subsection{Descartes Configurations and Curvature-Center Coordinates}  

736: 

737:  

738: In 1643, Descartes found a relation between the

739: radii for four mutually disjoint tangent circles of type (a)

740: in Figure \ref{fig1} below.

741: Let $r_1, r_2, r_3, r_4$ be the

742: radii of the tangent circles. 

743: Descartes showed~\footnote{Descartes expressed his relation in

744: a different form,  obtained by clearing denominators in \eqn{001}.}

745: a result equivalent to 

746: \beql{001}

747: \frac{1}{r_1^2}+\frac{1}{r_2^2}+\frac{1}{r_3^2}+\frac{1}{r_4^2}

748: =\frac{1}{2} \left(\frac{1}{r_1}+

749: \frac{1}{r_2}+\frac{1}{r_3} +\frac{1}{r_4}\right)^2,

750: \eeq

751: which is now called the {\em Descartes circle theorem}.

752: This result can be extended to apply to all Descartes

753: configurations, which can be of types (a)-(d) in

754: the figure below.

755: 

756: %

757: % Figure 2

758: %

759: 

760: \begin{figure}[htbp]

761: \centerline{\epsfxsize=3.0in \epsfbox{2-configs.ps}}

762: \caption{Descartes configurations}\label{fig1}

763: \end{figure}

764: 

765: In Figure \ref{fig1} (a) is a standard

766: Descartes  configuration in which the circles have disjoint interiors,

767: and all curvatures are positive.

768: We will also allow configurations like Figure~\ref{fig1} (b) 

769: in which one circle

770: encloses the other three.

771: We also allow straight lines to be

772: regarded as circles of curvature zero, leading to Descartes configurations

773: like Figure~\ref{fig1} (c) and (d), in which the ``interior'' 

774: of the ``circle''

775: defined by a straight line is defined to be a suitable half-plane.

776: 

777: 

778: To extend the Descartes Circle Theorem to all Descartes configurations

779: of types (a)-(d), we must define the curvatures to have

780: appropriate signs, as follows.  An {\em oriented circle} is a circle

781: together with an assigned direction of unit normal vector, which can

782: point inward or outward.  If it has radius $r$ then its {\em oriented

783: radius} is $r$ for an inward pointing normal and $-r$ for an outward

784: pointing normal.  Its {\em oriented curvature} (or ``signed

785: curvature'') is $1/r$ for an inward pointing normal and $-1/r$ for an

786: outward pointing normal.  By convention, the {\em interior} of an

787: oriented circle is its interior for an inward pointing normal and its

788: exterior for an outward pointing normal.  An {\em oriented Descartes

789: configuration} is a Descartes configuration in which the orientations

790: of the circles are compatible in the following sense:  either (i) the

791: interiors of all four oriented circles are disjoint, or (ii) the

792: interiors are disjoint when all orientations are reversed.  Each

793: Descartes configuration has exactly two compatible orientations in

794: this sense, one obtained from the other by reversing all orientations.

795: The {\em positive orientation} of a Descartes configuration is

796: the one in which the sum of the signed curvatures is positive, while

797: the {\em negative orientation} is the one in which the sum of

798: the curvatures is negative. (One can show 

799: that the sum of the signed curvatures is always nonzero.)  

800: Note that positive orientation corresponds to case (i),

801: and negative orientation to case (ii). With these

802: definitions, the Descartes circle theorem remains valid for all

803: oriented Descartes configurations, using oriented curvatures.

804: 

805: The geometry of Descartes configurations in Figure \ref{fig1}

806: is encoded in the curvature

807: vector $\bb=(b_1,b_2,b_3,b_4)^T$ where $b_i$ is the oriented curvature 

808: of the $i$th circle. If $\sum_{j = 1}^{4} b_j > 0$, then one of the

809: following holds:  (a) all of $b_1, ~b_2, b_3, b_4$ are positive;

810: (b) three are positive and one is negative; (c) three are positive

811: and one is zero; or (d) two are positive and equal and the other two

812: are zero. 

813: 

814: The Descartes circle theorem gives a quadratic equation

815: for the curvature vector $\bb$, which can be rewritten as

816: \beql{002}

817: \bb^T \bQ_{D} \bb = 0,

818: \eeq

819: in which

820: \beql{003}

821: \bQ_{D}= \bI - \frac{1}{2}{\bf 1}{\bf 1}^T =          

822:          \frac{1}{2}\left[ \begin{array}{rrrr}

823:                1 & -1 & -1 & -1 \\

824:                -1 & 1 & -1 & -1 \\

825:                -1 & -1 & 1 & -1 \\

826:                -1 & -1 & -1 & 1 \end{array} \right]   

827: \eeq

828: is the {\em Descartes quadratic form.} Here $\bo$ represents the vector

829: $(1,1,1,1)^T$. 

830: 

831: In \cite{LMW02} three of the authors of this paper

832: showed there exist matrix extensions of the

833: Descartes circle theorem which encode information

834: about both the curvatures and the centers of the

835: circles in the configuration, as follows.

836: Given an oriented circle $C$ with center $(x, y)$

837: and oriented curvature $b$  we define its

838: {\em curvature-center coordinates}  to be the $1 \times 3$

839: row vector $\bm(C) := (b, bx, by)$.

840: For the ``degenerate case'' of  an oriented  straight line $H$ we 

841: define its curvature-center coordinates as 

842: \[

843: \bm(H) :=(0, \bh),

844: \]

845: where $\bh=(h_1, h_2)$ is the unit normal vector giving the orientation of 

846: the straight line. 

847: %

848: %

849: % Theorem 3.1

850: %

851: %

852: 

853: 

854: \begin{theorem}[Extended Descartes Theorem]~\label{th31}

855: Given an ordered configuration $\sD$ 

856: of four oriented circles with 

857: oriented 

858: curvatures $(b_1,~b_2,~b_3,~b_4)$ and centers

859: $\{(x_i, y_i)~: 1 \leq i \leq 4\}$,

860: let $\bM_\sD$ be the $4 \times 3$ matrix

861: \beql{208}

862: \bM_\sD :=

863:         \left[ \begin{array}{ccl}

864:            b_1 & b_1x_1 & b_1y_1 \\

865:            b_2 & b_2x_2 & b_2y_2 \\

866:            b_3 & b_3x_3 & b_3y_3 \\

867:            b_4 & b_4x_4 & b_4y_4

868:         \end{array} \right] .

869: \eeq

870: We include the ``degenerate cases'' where some circles are

871: oriented lines. 

872: If this  configuration is an oriented Descartes configuration, then

873: $\bM=\bM_\sD$ has a nonzero first column and satisfies  

874: \beql{209}

875:         \bM^T \bQ_{D} \bM =

876:         \left[ \begin{array}{ccc}

877:            0 & 0 & 0 \\

878:            0 & 2 & 0 \\

879:            0 & 0 & 2

880:         \end{array} \right].

881: \eeq

882: Conversely, any real solution  $\bM$ to \eqn{209} 

883: with a nonzero first column is the 

884: curvature-center coordinate matrix $\bM_{\sD}$ of a unique 

885: ordered, oriented Descartes configuration $\sD$.  

886: \end{theorem}

887: 

888: \paragraph{Remark.} The 

889: hypothesis of a nonzero first column is

890: necessary in Theorem~\ref{th31}.

891: Given  the matrix $\bM$ of a Descartes configuration, the

892: matrix  $\tilde{\bM}$ obtained by zeroing out its first

893: column will continue to satisfy  \eqn{209}, and will not come from

894: a Descartes configuration.

895: 

896: 

897: \paragraph{Proof.}

898: An $n$-dimensional version of this theorem was 

899: stated as Theorem 3.2 in \cite{LMW02}, and the

900: ``if'' direction of the result was proved there. 

901: (The converse part of the theorem was not

902: proved there.) More precisely, in  \cite{LMW02} it was proved

903: that the curvature-center

904: coordinate matrix $\bM$

905: of any ordered, oriented  Descartes configuration

906: satisfies \eqn{209}.   The  first

907: column of the matrix $\bM$ of a Descartes

908: configuration cannot be identically zero because at

909: least two circles in any Descartes configuration have 

910: nonzero curvature.

911: 

912: To establish the  converse part of the theorem,

913: we need Theorem \ref{th32} below, which was

914: independently proved in \cite{LMW02}.  We postpone the proof 

915: to the end of \S3.1. 

916: $~~~\bsq$ 

917: 

918: 

919: 

920: %

921: % [maybe also in Part II \cite{GLMWY24}, maybe not]

922: % 

923: 

924: 

925: The  curvature-center coordinates $\bm(C)$  uniquely determine

926: a circle in general position, but they fail to be unique

927: for the degenerate case of a line, where the information

928: $(0, \bh)$ determines only a  direction normal to the line

929: and not its position.

930: 

931: In  \cite{LMW02} three of the authors defined

932: {\em augmented curvature-center coordinates},

933: (called {\em ACC-coordinates} for short), 

934: $\bw(C)$ of an

935: oriented circle $C$ with signed curvature $b$

936: and center $(x,y)$, which resolve this ambiguity.

937:  These represent $C$ by a row vector

938: \beql{ACC}

939: \bw(C) := (\bar{b} ,  b , b x_1, bx_2) 

940: \eeq

941: in which $\bar{b}$ is the signed curvature of the oriented circle

942: obtained by inversion in the unit circle.

943: The operation of inversion in the unit circle acts on $\RR^2$ by

944: $$

945: (x, y) \mapsto (x^{'}, y^{'}) = ( \frac{x}{x^2 + y^2}, \frac{y}{x^2 + y^2}).

946: $$

947: It maps a circle $C$ of finite (oriented) radius $r$ to the circle 

948: $\bar{C}$ with center $\bar{\bx} =

949: {\bx}/(|\bx|^2 - r^2 )$ and oriented radius 

950: $\bar{r} = \frac{r}{|\bx|^2 -r^2}$, 

951: having  oriented curvature  $\bar{b}= \frac{|\bx|^2 -r^2}{r}$.

952: If $C$ is a straight line with specified normal

953: direction, we determine $\bar{C}$ as the

954: image of this line, with orientation coming from the 

955: specified normal. In the degenerate case that $\bar{C}$ is a 

956: straight line we define

957: $\bar{b}=0$. In all cases,

958: \beql{303f}

959: (b x, b y) =  \frac {\bx}{r} = \frac{\bar{\bx}}{\bar{r}} = 

960: (\bar{b} x', \bar{b} y'),

961: \eeq

962: so that $C$ and $\bar{C}$ have the same curvature$\times$center data.

963: 

964: Augmented curvature-center coordinates provide a global coordinate

965: system:  no two distinct oriented circles have the same coordinates.

966: The only case to resolve is when $C$ is a straight line, i.e., $b=0$.

967: The relation \eqn{303f} shows that $(\bar{b}, b x,b y)$

968: are the curvature-center coordinates of $\bar{C}$, and if $\bar{b}

969: \neq 0$, this uniquely determines $\bar{C}$; inversion in the unit

970: circle then determines $C$. In fact, $\bar{b}$ in this case is twice

971: the distance of $C$ from the origin. In the remaining case, $b = \bar{b} =0$

972: and $C = \bar{C}$ is the unique line passing through the origin

973: whose unit normal is given by the remaining coordinates.

974: 

975: Given a collection $(C_1, C_2, C_3, C_4)$ of four oriented

976: circles (possibly lines) in $\rr^2$, the {\em augmented matrix}

977: $\bW$ associated with it is the $4 \times 4$ matrix whose

978: $j$-th row has entries given by the augmented curvature-center

979: coordinates $\bw(C_j)$ of the $j$-th circle.

980: The following result characterizes oriented

981: Descartes configurations.

982: 

983: %

984: %

985: % Theorem 3.2

986: %

987: %

988: 

989: \begin{theorem}[Augmented Euclidean Descartes Theorem]~\label{th32}

990: Given an ordered configuration $\sD$ of four oriented circles (or lines) 

991: $\{C_i:  1 \leq i \leq 4\}$

992: with 

993: curvatures $(b_1,~b_2,~b_3,~b_4)$ and centers

994: $\{(x_i, y_i)~: 1 \leq i \leq 4\}$,

995: let $\bW_\sD$ be the $4 \times 4$ matrix

996: \beql{308}

997: \bW_\sD :=

998:         \left[ \begin{array}{rrrr}

999:          \bar{b}_1  &  b_1 & b_1x_1 & b_1y_1 \\

1000:          \bar{b}_2  &  b_2 & b_2x_2 & b_2y_2 \\

1001:          \bar{b}_3  &  b_3 & b_3x_3 & b_3y_3 \\

1002:          \bar{b}_4  &  b_4 & b_4x_4 & b_4y_4

1003:         \end{array} \right] .

1004: \eeq

1005: If $\sD$ is an (ordered, oriented)  

1006: Descartes configuration then $\bW_{\sD}$ satisfies 

1007: \beql{307}

1008:         \bW^T \bQ_{D}  \bW = 

1009:         \left[ \begin{array}{rrrr}

1010:            0 & -4 & 0  & 0 \\

1011:           -4 &  0 & 0  & 0 \\

1012:            0 &  0 & 2  & 0 \\

1013:            0 &  0 & 0  & 2 \\ \end{array} \right].     

1014: \eeq

1015: Conversely, any real solution $\bW$ to \eqn{307} is the augmented

1016: matrix $\bW_{\sD}$  of a unique ordered,

1017: oriented Descartes configuration $\sD$.  

1018: \end{theorem}

1019: 

1020: \paragraph{Proof.} This is proved as the

1021: two-dimensional case of  Theorem 3.3  in \cite{LMW02}. $~~~\bsq$ \\

1022: 

1023: We call the quadratic form $Q_W$ defined by the matrix 

1024: \beql{308b}

1025: \bQ_{W}= \left[ \begin{array}{rrrr}

1026:            0 & -4 & 0  & 0 \\

1027:           -4 &  0 & 0  & 0 \\

1028:            0 &  0 & 2  & 0 \\

1029:            0 &  0 & 0  & 2 \\ \end{array} \right]      

1030: \eeq

1031: the {\em Wilker quadratic form}.

1032: We name this quadratic form after  J. B. Wilker ~\cite{Wi81}, who

1033: introduced in spherical geometry a coordinate system

1034: analogous to augmented curvature-center coordinates, see

1035: \cite[\S2 p. 388-390 and \S9]{Wi81}. However Wilker did not

1036: explicitly formulate any result  exhibiting the quadratic form

1037: $Q_{W}$, see  \cite[p. 349 ]{LMW02}. 

1038: 

1039: Theorem~\ref{th32}

1040: identifies the set of all ordered, oriented

1041: Descartes configurations $\DD$  with the set $\sM_{\DD}$ of real   

1042: solutions $\bW = \bW_{\sD}$  to the matrix equation \eqn{307}.

1043: This equation 

1044: states that the  augmented matrix coordinates

1045: of an oriented Descartes configuration give an intertwining map

1046: between the Descartes form and the Wilker form.

1047: The set $\sM_{\DD}$ has the structure of a six-dimensional

1048: affine real-algebraic variety.

1049: 

1050: It is possible to refine the parameter space

1051: to  a {\em moduli space} 

1052: $\tilde{\sM}_{\DD}$

1053: of (unordered, unoriented) Descartes configurations as an

1054: orbifold $\sM_{\DD}/\sim$ obtained by quotienting by

1055: a finite group of order $48$ (acting on the left). This group

1056: is generated by the $4\times 4$ permutation

1057: matrices (permuting rows) and $-\bI$, which reverses total

1058: orientation.  This orbifold has singular points at those

1059: ordered, oriented Descartes configurations which remain  

1060: invariant under a non-trivial permutation matrix.

1061: For our purposes it is more convenient to use 

1062: the parameter space $\sM_{\DD}$ which is a smooth manifold.

1063: 

1064: 

1065: Both the Descartes quadratic form and the Wilker quadratic form

1066: are equivalent over the real numbers to the Lorentz

1067: quadratic form 

1068: $Q_{\sL}(x) := - x_0^2 + x_1^2 + x_2^3 + x_{3}^2,$

1069: with associated matrix

1070: \beql{308c}

1071: \bQ_{\sL}= \left[

1072: \begin{array}{crrr}

1073: -1 & 0 & 0 & 0 \\

1074:  0 & 1 & 0 & 0 \\

1075:  0 & 0 & 1 & 0 \\

1076:  0 & 0 & 0 & 1

1077: \end{array}

1078: \right].

1079: \eeq

1080: For any quadratic form $Q$, let  $Aut(Q)$  be the group of automorphisms

1081: under the ``congruence action'', defined by 

1082: $$

1083: Aut(Q)=\{ \bU  \in GL(4, \RR) : \bU^T \bQ \bU = \bQ \},

1084: $$

1085: where $\bQ$ is the symmetric matrix that represents $Q$. 

1086: The Lorentz quadratic form has a large group of automorphisms

1087: $Aut(Q_{\sL})$, 

1088: which is exactly the real Lorentz group $O(3, 1)$.

1089: 

1090: The Descartes form and Wilker form are  not only equivalent

1091: to the Lorentz form  over the real numbers,

1092: but also  over the rational numbers. For the Descartes

1093: form one  has

1094: \beql{309aa}

1095: \bQ_{D} = \bJ_0^T \bQ_{\sL} \bJ_0,

1096: \eeq

1097: in which

1098: \begin{equation}~\label{309}

1099: \bJ_0 = \frac{1}{2} \left[

1100: \begin{array}{crrr}

1101: 1 & 1 & 1 & 1 \\

1102: 1 & 1 & -1 & -1 \\

1103: 1 & -1 & 1 & -1 \\

1104: 1 & -1 & -1 & 1

1105: \end{array}

1106: \right]

1107: \end{equation}

1108: and $\bJ_0 = \bJ_0^T = \bJ_0^{-1}$.

1109: It follows that

1110: \beql{N309c}

1111: Aut(Q_{D})=\bJ_{0}^{-1} O(3,1) \bJ_0.

1112: \eeq

1113: 

1114: 

1115: 

1116: 

1117: 

1118: The rational equivalence

1119: of the Wilker quadratic form $Q_{W}$ 

1120: to the Lorentz form $Q_{\sL}$ follows from Theorem~\ref{th32},

1121: as soon as we exhibit a  Descartes configuration $\sD$

1122: whose augmented curvature-center coordinates $\bW_{\sD}$ are

1123: a rational matrix. One is given by

1124: \begin{equation}~\label{N330a}

1125: \bW_0= \bW_{\sD}= \left[

1126: \begin{array}{rrrr}

1127:  2 & 0 & 0 & 1 \\

1128:  2 & 0 & 0 & -1 \\

1129:  0 & 1 & 1 & 0 \\

1130:  0 & 1 & -1 & 0

1131: \end{array}

1132: \right].

1133: \end{equation}

1134: It corresponds to the positively oriented Descartes configuration pictured in

1135: Figure \ref{fig2.1a}, in which the dotted line is the $x$-axis

1136: and the two circles touch at the origin.

1137: %

1138: %

1139: % Figure 3: strongly integral

1140: %

1141: %

1142: 

1143: \begin{figure}[htbp]

1144: \centerline{\epsfxsize=3.0in \epsfbox{3-strongly.ps}}

1145: \caption{A strongly integral Descartes configuration}~\label{fig2.1a}

1146: \end{figure}

1147: 

1148: We then have 

1149: \begin{equation}~\label{N333a}

1150: \bQ_{W} = \bA^{T} \bQ_{\sL} \bA,  

1151: \end{equation}

1152: with 

1153: \begin{equation}~\label{N333b}

1154: \bA =  \bJ_0 \bW_0=\left[

1155: \begin{array}{crrr}

1156:  2 & 1 &  0 & 0 \\

1157:  2 & -1 & 0 & 0 \\

1158:  0 & 0 & 1 &  1 \\

1159:  0 & 0 & -1 & 1

1160: \end{array}

1161: \right]. 

1162: \end{equation}

1163: It follows that

1164: \beql{N333c}

1165: Aut(Q_W)=\bA^{-1} O(3,1) \bA.

1166: \eeq

1167: 

1168: Finally we note that the determinants of these quadratic

1169: forms are given by  

1170: \beql{Ndet1}

1171: \det( \bQ_{D}) = \det(\bQ_{\sL}) = -1~~\mbox{and}~~ \det(\bQ_{W}) = -64.

1172: \eeq

1173: Taking determinants in  \eqn{307} yields 

1174: that augmented curvature-center coordinate

1175: matrices $\bW_{\sD}$ have

1176: \beql{Ndet2}

1177: \det( \bW_{\sD}) = \pm 8. 

1178: \eeq

1179: Examples show that  both  values $\pm 8$ occur.

1180: 

1181: %

1182: % ``only if'' proof of Thm. 3.1

1183: %

1184: \vspace{.3cm} 

1185: \noindent {\bf Proof of the ``only if'' part of Theorem \ref{th31}.} 

1186: We note that if $\bM$ is the curvature-center coordinate

1187: matrix of a Descartes configuration, this configuration

1188: is unique. Indeed at least two circles in the configuration

1189: have nonzero curvature, so are determined by their

1190: curvature-center coordinates. This gives enough information

1191: to find the location of any lines in the configuration,

1192: and their signed normal vectors  are determined uniquely

1193: by their curvature-center coordinates.

1194: Thus it  suffices to show that there exists a $4 \times 4$ matrix $\bW$

1195: whose last three columns agree with $\bM$, such that

1196: \begin{equation}~\label{A1A}

1197: \bW^T \bQ_{D} \bW = \bQ_{W} = \left[ \begin{array}{rrrr}

1198:            0 & -4 & 0  & 0 \\

1199:           -4 &  0 & 0  & 0 \\

1200:            0 &  0 & 2  & 0 \\

1201:            0 &  0 & 0  & 2 \\ \end{array} \right].

1202: \end{equation}

1203: If so,  then Theorem \ref{th32}  implies

1204: that $\bW$ is the augmented curvature-center coordinate

1205: matrix of a unique ordered, oriented Descartes configuration $\sD$.

1206: It follows that $\bM$ is the curvature-center coordinate matrix

1207: of $\sD$.

1208: 

1209: This discussion implies that the extension $\bW$ is unique,

1210: and we proceed to find it by transforming the problem

1211: from the Descartes form $\bQ_{D}$ to the Lorentz form

1212: $\bQ_{\sL} = \bJ_0^T \bQ_{D} \bJ_0$.  This proof

1213: seems ``pulled out of thin air'' in that it  guesses to do

1214: the ``translation'' by the matrix $\bT$ below, in terms of

1215: which the extended matrix $\tilde{\bW}$ has a simple form.

1216: We set

1217: $\tilde{\bM} = \bJ_0 \bM$

1218: with $\bJ_0$ given by \eqn{309}, so that  $\bJ_0 = \bJ_0^T = \bJ_0^{-1}$.

1219: Then

1220: \beql{A2}

1221: \tilde{\bM}^T \bQ_{\sL}\tilde{\bM} = \bM^T \bQ_{D} \bM =  \bQ_{0} :=

1222:  \left[

1223: \begin{array}{ccc}

1224:            0 & 0 & 0 \\

1225:            0 & 2 & 0 \\

1226:            0 & 0 & 2

1227:         \end{array} \right].

1228: \eeq

1229: We now note that

1230: $$

1231: \tilde{\bM}_{11} = \frac{1}{2}( m_{11} + m_{12} + m_{13} + m_{14}) \ne 0.

1232: $$

1233: To verify this, note that the top left entry of  \eqn{209} gives

1234: $$

1235: Q_{D}(m_{11},  m_{12}, m_{13}, m_{14}) =

1236: (m_{11}^2 + m_{12}^2 + m_{13}^2+ m_{14}^2 ) -

1237: \frac{1}{2} ( m_{11} + m_{12} + m_{13} + m_{14})^2 = 0,

1238: $$

1239: and if $m_{11} + m_{12} + m_{13} + m_{14}= 0$ then we would obtain

1240: $m_{11}^2 + m_{12}^2 + m_{13}^2+ m_{14}^2=0$, which forces

1241: all  $m_{1j}=0$, contradicting the hypothesis.

1242: Now there exists a unique matrix

1243: $$

1244: \bT= \left[

1245: \begin{array}{ccc}

1246:  a & b & c \\

1247:  0 & 1 & 0 \\

1248:  0 & 0 & 1

1249: \end{array} \right]

1250: $$

1251: such that the first row of $\tilde{\bM} \bT$ is $[ 1, 0, 0]$,

1252: which takes $a = (\tilde{\bM}_{11})^{-1}$,  so that $\bT$ is invertible.

1253: We write

1254: $$

1255: \tilde{\bM} \bT = \left[

1256: \begin{array}{ccc}

1257:  1 & 0 & 0 \\

1258:  v_{21} & v_{22} \sqrt{2} & v_{23} \sqrt{2} \\

1259:  v_{31} & v_{32} \sqrt{2} & v_{33} \sqrt{2} \\

1260:  v_{41} & v_{42} \sqrt{2} & v_{43} \sqrt{2}

1261: \end{array}

1262: \right] ~,

1263: $$

1264: and then have

1265: \begin{equation}\label{A3}

1266: \bT^T \tilde{\bM}^T \bQ_{\sL} \tilde{\bM} \bT =

1267: \bT^T  \bQ_0 \bT = \bQ_0 = \left[

1268:        \begin{array}{ccc}

1269:            0 & 0 & 0 \\

1270:            0 & 2 & 0 \\

1271:            0 & 0 & 2

1272:         \end{array}

1273: \right].

1274: \end{equation}

1275: This  matrix  equation is equivalent to the assertion that

1276: \begin{equation*}

1277: \bV := \left[

1278: \begin{array}{ccc}

1279: v_{21} & v_{22} & v_{23} \\

1280: v_{31} & v_{32} & v_{33} \\

1281: v_{41} & v_{42} & v_{43}

1282: \end{array} \right]

1283: \end{equation*}

1284: is an orthogonal matrix, i.e. $\bV^T \bV = \bV \bV^T =I.$

1285: We now define

1286: $$

1287: \tilde{\bW} :=

1288: \left[

1289: \begin{array}{cccc}

1290: 2 &  1 & 0 & 0 \\

1291: -2 v_{21} &  v_{21} & v_{22} \sqrt{2} & v_{23} \sqrt{2} \\

1292: -2 v_{31} &  v_{31} & v_{32} \sqrt{2} & v_{33} \sqrt{2} \\

1293: -2 v_{41} &  v_{41} & v_{42} \sqrt{2} & v_{43} \sqrt{2}

1294: \end{array}

1295: \right],

1296: $$

1297: whose last three columns match $\tilde{\bM} \bT$,

1298: and verify by direct calculation using \eqn{A3}  that

1299: $$

1300: \tilde{\bW}^T \bQ_{\sL} \tilde{\bW} = \bQ_{W}.

1301: $$

1302: We define $\bW := \bJ_0^{-1} \tilde{\bW} = \bJ_0 \tilde{\bW}$,

1303: and find that

1304: $$

1305: \bW^T \bQ_{D} \bW = \tilde{\bW}^T \bJ_0  \bQ_{\sL} \bJ_0 \tilde{\bW} = \bQ_{W}.

1306: $$

1307: This is the desired lift,

1308: since  the last three columns of $\bW$ are exactly $\bM$.$~~~\bsq$

1309: 

1310: 

1311: 

1312: 

1313: 

1314: 

1315: 

1316: 

1317:  

1318: 

1319: 

1320: %*****************************************************************************

1321: %

1322: % SECTION 3.2 

1323: %

1324: %*****************************************************************************

1325: 

1326: \subsection{M\"{o}bius and Lorentz group actions}

1327: 

1328: 

1329: The augmented Euclidean Descartes theorem immediately 

1330: yields two group actions on the space $\sM_{\DD}$

1331: of ordered, oriented Descartes configurations.

1332: The group $Aut( Q_{D})$ acts on the left

1333: and the group $Aut(Q_{W})$ acts on the right, as

1334: $$

1335: \bW_{\sD}  \mapsto \bU  \bW_{\sD}  \bV^{-1}, ~~\mbox{with}~~ 

1336: \bU \in Aut( Q_{D}),~ 

1337: \bV \in Aut(Q_{W}).

1338: $$

1339: The two group actions clearly commute with each other.

1340: Both groups are  conjugate to the real Lorentz group

1341: $O(3, 1)$, and therefore they each have four connected 

1342: components. These components are specified for any  $\bY \in O(3,1)$

1343: by the

1344: sign of $\det(\bY)$,  and by the sign of ``total orientation'', 

1345: which is the sign of  

1346: $\bY_{11} = \be^T_1 \bY \be_1$, in which $\be^T_1=(1,0,0,0)$. 

1347: For  $\bU \in Aut(Q_{D})$, it is the sign of $\bo^T \bU \bo$, in which 

1348: $\bo^T=(1,1,1,1)$; and for $\bV\in Aut(Q_W)$, it is the sign of 

1349: $\be^T_1 \bA \bV \bA^{-1} \be_1 $ in which  $\bA$ is any matrix satisfying 

1350: $O(3,1)=\bA  Aut(Q_W)\bA^{-1}$, such as \eqn{N333b}.  

1351: 

1352: The parameter space $\sM_{\DD}$ also has four connected

1353: components, specified by similar invariants, which are the sign

1354: of $\det(\bW)$ and the {\em total orientation}, which is the sign of the

1355: sum\footnote{The sum of the signed curvatures of the circles 

1356: of an oriented  Descartes configuration cannot

1357: be zero.}  

1358: of the (signed) curvatures of the four circles in the

1359: Descartes configuration.

1360: We let

1361: $$

1362: \sM_{\DD} = \sM_{+}^{\uparrow} \cup \sM_{-}^{\uparrow} \cup 

1363: \sM_{+}^{\downarrow} \cup \sM_{-}^{\downarrow}, 

1364: $$

1365: in which the subscript describes the sign of the determinant and

1366: the superscript the orientation, with $\uparrow$ being 

1367: positive orientation. We also let

1368: $$

1369: \sM_{\DD}^{\uparrow} := \sM_{+}^{\uparrow} \cup \sM_{-}^{\uparrow} 

1370: $$

1371: denote the set of positively oriented Descartes configurations.

1372: The relevance of this decomposition is that the Apollonian group

1373: defined in \S4 leaves the set of positively oriented Descartes

1374: configurations 

1375: $\sM_{\DD}^{\uparrow}$ invariant.

1376: 

1377: 

1378: The action on the right by the elements of $Aut(Q_{W})$

1379: maps circles to circles, since the 

1380: circles in a Descartes configuration correspond 

1381: to the rows in the matrix $\bW_{\sD}$ of an ordered, oriented

1382: Descartes configuration.  

1383: This right action 

1384: can essentially be identified

1385: with the general M\"{o}bius group M\"{o}b(2) generated

1386: by the  linear fractional

1387: transformations acting on the one-point compactification

1388: $\hat{\RR}^2$ of $\RR^2$, which is $PSL(2, \CC) = SL(2, \CC)/\{ \pm \bI\}$,

1389: together with  complex conjugation $\bz \mapsto \bar{\bz}$.

1390: and has two connected components.

1391: More precisely,

1392: $Aut(Q_W)$ is isomorphic to 

1393: the direct sum of the M\"{o}bius group with a group of

1394: order two, which has four connected components,

1395: as explained in Appendix A.

1396: 

1397: 

1398: The action on the left, by $Aut(Q_{D})$,

1399: produces a new oriented Descartes configuration

1400: whose coordinates  mix together the coordinates of the 

1401: different circles in the original   Descartes configuration.

1402: This group action  does

1403: not make sense as an action on individual circles

1404: in the configuration.

1405: This group action is intrinsically associated to

1406: the $6$-dimensional space of oriented

1407: Descartes configurations.

1408: 

1409: \begin{theorem}~\label{th33}

1410: (1) The groups $Aut(Q_{D})$ and $Aut(Q_{W})$ are

1411: conjugate to $Aut(Q_{\sL}) \equiv  O(3,1)$.

1412: 

1413: (2) The group $Aut(Q_{D})$ acts transitively on the 

1414: left on the space $\sM_{\DD}$

1415: of all ordered, oriented  Descartes configurations. 

1416: Given two such 

1417:  Descartes configurations $\sD$ and $\sD^{'}$

1418: there exists a unique $\bU \in Aut(Q_{D})$ such that

1419: $\bU \bW_{\sD} = \bW_{\sD'}$.

1420: 

1421: (3) The group $Aut(Q_{W})$ acts transitively on the 

1422: right on the space 

1423: of all ordered, oriented  Descartes configurations $\sM_{\DD}$. 

1424: Given two such Descartes configurations $\sD$ and $\sD^{'}$

1425: there exists a unique $\bV \in Aut(Q_{W})$ such that

1426: $\bW_{\sD} \bV^{-1} = \bW_{\sD'}$.

1427: 

1428: 

1429: (4) The action of $Aut(Q_{D})$ on the space 

1430: $\sM_{\DD}$ commutes with  the action of $Aut(Q_{W})$. 

1431: \end{theorem}

1432: 

1433: \paragraph{Remark.} The left action by $Aut(Q_{D})$

1434: and the right action by $Aut(Q_W)$ on $\sM_{\DD}$

1435: given by Theorem \ref{th33} can be identified with a left and right

1436: action of the Lorentz group $O(3,1) = Aut(Q_{\sL})$ on $\sM_{\DD},$

1437: using \eqn{N309c} and \eqn{N333c}, respectively.

1438: Theorem~\ref{th33} 

1439: shows that both these actions are  transitive, and that  the

1440: stabilizer of a point is the identity element.

1441: This is equivalent to saying that 

1442: the space $\sM_{\DD}$ is a {\em principal homogeneous space} 

1443: (or {\em torsor}) for 

1444: $O(3,1)$ for either action.

1445: 

1446: \paragraph{Proof.}

1447: Part (1) follows from the conjugacy between the

1448: Descartes and Wilker forms and the Lorentz form $Q_{\sL}$

1449: given in \eqn{309aa} and \eqn{N333a}, respectively.

1450: These give

1451: $$

1452: Aut(Q_{D}) = \bJ_0^{-1} Aut(Q_\sL) \bJ_0 \equiv \bJ_0^{-1} O(3,1) \bJ_0,

1453: $$

1454: and

1455: $$

1456: Aut(Q_{W}) = \bA^{-1} Aut(Q_{\sL}) \bA \equiv \bA^{-1} O(3,1) \bA.

1457: $$

1458:  

1459: 

1460: Parts (2) and (3) follow immediately from (1). Given a fixed

1461: $\bW= \bW_{\sD} \in \sM_{\dd}$, we assert that

1462: \begin{equation}~\label{N335}

1463: \sM_{\DD} = Aut(Q_{D}) \bW. 

1464: \end{equation}

1465: Taking  $\bW_0$ in \eqn{N330a} we have

1466: $$

1467: (\bW \bW_0^{-1})^T \bQ_{D} (\bW \bW_0^{-1})=

1468: (\bW \bA^{-1}\bJ_0)^{T} \bQ_{D} (\bW \bA^{-1}\bJ_0) = \bQ_{D},

1469: $$

1470: so that $\bW \bW_0^{-1} \in Aut(Q_{D})$ and

1471:  $Aut(Q_{D}) \bW \bW_0^{-1}$ forms  a single orbit of $Aut(Q_{D})$,

1472: The map $\bW_{\sD} \mapsto \bW_{\sD} \bW_0^{-1}$ from 

1473: $\sM_{\dd}$ into $Aut(Q_{D})$ is one-to-one since $\bW_0$

1474: is invertible, and it is onto since the domain includes

1475: $Aut(Q_{D}) \bW$. Thus 

1476: \eqn{N335} follows, and this 

1477: gives (2).  We similarly obtain 

1478: $$

1479: \sM_{\DD} = \bW Aut(Q_{W})^{-1} = \bW Aut(Q_{W}),

1480: $$

1481: which gives (3).

1482: 

1483: Finally, part (4) follows directly from parts (2) and (3),

1484: since in general a left and a right action of two matrix groups

1485: on a space of matrices commute.

1486: ~~~$\bsq$ \\

1487: 

1488: 

1489: In this paper we mainly study  structures 

1490: associated to the left action of

1491: $Aut(Q_{D})$ on $\sM_{\DD}$.  We term the left action the 

1492: {\em Lorentz action}, although this  is a misnomer, since the right

1493: action can also be identified with a Lorentz group action. 

1494: However we wish to assign different names for the two actions, 

1495: and the right

1496: action is conveniently called

1497: the {\em M\"{o}bius action},

1498: because of its relation to a 

1499: M\"{o}bius group action on circles on the Riemann sphere,

1500: detailed in Appendix A. 

1501: 

1502: 

1503: 

1504: %*****************************************************************************

1505: %

1506: % SECTION 3.3 

1507: %

1508: %*****************************************************************************

1509: 

1510: \subsection{Distinguished Elements of $Aut(Q_{D})$.}

1511: 

1512: %In this paper we are concerned with the action of

1513: %certain elements of 

1514: %$Aut(Q_{D})$ on Descartes configurations. 

1515: We now describe 

1516: some  specific elements of $Aut(Q_{D})$ that have a 

1517: nice geometrically

1518: interpretable action on every Descartes configuration,

1519: visualizable in terms of inversions, and whose 

1520: associated matrices have integer (or half-integer) entries.

1521: 

1522: 

1523: The first set of four operations correspond to inversion in 

1524: the circle determined by the three intersection points  of circles

1525: in a Descartes configuration $\sD$ that avoid  one 

1526: particular circle. There are four possibilities for these.

1527: This inversion fixes the three circles involved in

1528: the intersections and moves the fourth circle, to

1529: the unique other circle that is tangent to the first

1530: three circles. 

1531: For this reason we call it the  {\em reflection operator}. 

1532: Let 

1533: $\fs_1=\fs_1[\sD]$ denote the M\"{o}bius

1534: transformation  of this kind that moves the 

1535: circle $C_1$. In particular, $\fs_1$ maps the Descartes 

1536: configuration $\sD=(C_1, C_2, C_3, C_4)$ to $\fs_1(\sD)=(C_1', C_2, 

1537: C_3, C_4)$.  

1538: This reflection operator depends on the specific Descartes configuration.

1539: However for  all Descartes configurations $\sD$ there holds 

1540: \beql{N331}

1541:   \bW_{\fs_1(\sD)}=  \bS_1 \bW_{\sD}  ,

1542: \eeq

1543: where  $\bS_1 \in Aut(Q_{D})$ is given by 

1544: $$

1545: \bS_1  =  \left[

1546: \begin{array}{rccc}

1547: -1 & 2 & 2 & 2 \\

1548: 0 & 1 & 0 & 0 \\ 

1549: 0 & 0 & 1 & 0 \\

1550: 0 & 0 & 0 & 1

1551: \end{array} \right]

1552: $$

1553: belongs to $Aut(Q_{D})$ and is independent of the location of $\sD$.

1554: The geometric action of $\fs_1$  is pictured in Figure~\ref{fig4}.

1555: The other three operations $\fs_2, \fs_3, \fs_4$ give similar matrices

1556: $\bS_2, \bS_3, \bS_4$, 

1557: obtained by permuting the first and j-th rows and columns of $\bS_1$

1558: for $2 \le j \le 4$.

1559: 

1560: %

1561: % fig 4

1562: %

1563: %

1564: 

1565: \begin{figure}[htbp]

1566: \centerline{\epsfxsize=3.0in \epsfbox{4-reflection.ps}}

1567: \caption{The reflection operator $\fs_1$ }~\label{fig4}

1568: \end{figure}

1569: 

1570: To prove \eqn{N331} we check  that 

1571: $\bS_1 \in Aut(Q_{D})$. 

1572: Then for any Descartes

1573: configuration $\sD$  Theorem~\ref{th32} gives 

1574: that  $\bS_1 \bW_{\sD} =\bW_{\sD'}$ for some

1575: Descartes configuration $\sD'$.  The configuration

1576: $\sD'$ necessarily has three oriented circles fixed, with the fourth

1577: moved, so the fourth one must be the unique

1578: other circle tangent to the given three, and its orientation

1579: is uniquely determined by the other three orientations. Now

1580: $\bW_{\fs_1(\sD)}$ corresponds to  that Descartes

1581: configuration  consisting  of the same four ordered circles,

1582: which has the same (positive or negative) orientation as $\sD'$. 

1583: 

1584: A second set of four operations corresponds to inversion in

1585: one of the four circles in a Descartes configuration.

1586: Let that circle be $C_1$. Now $C_1$ remains fixed,

1587: while the other three circles change. Denoting

1588: this inversion by $\fs_1^{\perp}$,  it is easy to show that

1589: \beql{N323a}

1590: \bW_{\fs_1^{\perp}(\sD)} = \bS_1^{\perp} \bW_{\sD},

1591: \eeq

1592: where $\bS_1^{\perp} \in Aut(Q_{D})$ is given by

1593: $$

1594: \bS_1^{\perp} =  \left[

1595: \begin{array}{rccc}

1596: -1 & 0 & 0 & 0 \\

1597: 2 & 1 & 0 & 0 \\

1598: 2 & 0 & 1 & 0 \\

1599: 2 & 0 & 0 & 1

1600: \end{array} \right].

1601: $$

1602: The corresponding matrices $\bS_2^{\perp},\bS_3^{\perp},\bS_4^{\perp}$

1603: are obtained by permuting the first and j-th rows and columns of

1604: $\bS_1^{\perp}$ for $2 \le j \le 4$.

1605: The operation $\fs_1^{\perp}$ is pictured in Figure~\ref{fig5}.

1606: 

1607: %

1608: % figure 5

1609: %

1610: 

1611: \begin{figure}[htbp]

1612: \centerline{\epsfxsize=3.0in \epsfbox{5-inversion.ps}}

1613: \caption{The inversion operation $\fs_1^\perp$ }~\label{fig5}

1614: \end{figure}

1615: 

1616: 

1617: Finally we describe an operator, 

1618: which we call the {\em duality operator} $\bD$,  

1619: and which is an involution, as follows.

1620: Given a Descartes configuration $\sD$, there exists a

1621: Descartes configuration $\sD'$ obtained from it which

1622: consists of the four circles each of which passes through

1623: the three intersection points avoiding one circle.

1624: The resulting configuration has the same six points of

1625: tangency as the original configuration, and the circles

1626: in the configuration $\sD'$ are perpendicular to the

1627: circles of $\sD$ at these tangency points. This is

1628: pictured in Figure~\ref{dualfig}.

1629: 

1630: %

1631: % fiigure 6

1632: %

1633: 

1634: \begin{figure}[htbp]

1635: \centerline{\epsfxsize=2.0in \epsfbox{6-dual.ps}}

1636: \caption{The dual operation}~\label{dualfig}

1637: \end{figure}

1638: 

1639: Straightforward  computation shows that

1640: $$

1641: \bW_{\sD'}  = \bD \bW_{\sD},

1642: $$

1643: with

1644: $$

1645: \bD = \frac{1}{2} \left[

1646: \begin{array}{crrr}

1647: -1 & 1 & 1 &  1 \\

1648: 1 & -1 &  1 &  1 \\

1649: 1 &  1 & -1 &  1 \\

1650: 1 & 1 &  1 & -1

1651: \end{array}

1652: \right].

1653: $$

1654: Note that  $\bD = -\bQ_{D} \in Aut (Q_{D})$. There is

1655: a M\"{o}bius transformation $\fd = \fd_{\sD}$ which

1656: sends $\sD \mapsto \fd(\sD) =  \sD'$, and  it

1657: depends on $\sD$.

1658: 

1659: %*****************************************************************************

1660: %

1661: % SECTION 4. 

1662: %

1663: %*****************************************************************************

1664: %

1665: 

1666: \section{Apollonian Packings and the Apollonian Group}

1667: \setcounter{equation}{0}

1668: 

1669: 

1670: Apollonian circle packings are infinite packings of circles

1671: recursively constructed from a given 

1672: positively oriented Descartes configuration $\sD$.

1673: For simplicity consider a

1674: positively oriented  Descartes configuration

1675: $\sD = \{C_1, C_2, C_3, C_4\}$ in which circle $C_4$ encloses the other three,

1676: so that the interior of $C_4$ includes the point $\bz_\infty$ at infinity.

1677: The uncovered area consists of four lunes, indicated by the shading 

1678: in Figure~\ref{fig3.1}.

1679: 

1680: %

1681: % figure 7

1682: %

1683: 

1684: \begin{figure}[htbp]

1685: \centerline{\epsfxsize=2.0in \epsfbox{7-stagezero.ps}}

1686: \caption{A zero-th stage packing $\sP_{\sD}^{(0)}$ 

1687: and uncovered regions (lunes)}\label{fig3.1}

1688: \end{figure}

1689: 

1690: The zeroth stage packing $\sP_{\sD}^{(0)} = \sD$.

1691: At the first stage we inscribe a circle in each lune,

1692: to obtain a circle packing $\sP_{\sD}^{(1)}$ containing 8 circles.

1693: Each of these circles lies in a unique (unordered) Descartes configuration

1694: in the first stage packing, consisting of it and the three circles it touches.

1695: The uncovered area that remains consists of 12 lunes. See Figure \ref{fig3.2}.

1696: 

1697: %

1698: % figure 8

1699: %

1700:  

1701: \begin{figure}[htbp]

1702: \centerline{\epsfxsize=2.0in \epsfbox{8-stageone.ps}}

1703: \caption{A first  stage packing $\sP_{\sD}^{(1)}$ }\label{fig3.2}

1704: \end{figure}

1705: 

1706: 

1707: 

1708: At the second stage we inscribe a circle in each of these lunes, which 

1709: produces the second stage packing $\sP_{\sD}^{(2)}$.

1710: Continuing in this way, at the $n$-th stage we add $4 \cdot 3^{n-1}$ circles, 

1711: and each of these lies in a unique

1712: Descartes configuration in the $n$-th stage partial packing.

1713: The {\em Apollonian packing} $\sP_{\sD}$ associated to $\sD$ is

1714: the limit packing 

1715: \beql{T31}

1716: \sP_{\sD} := \bigcup_{n=1}^\infty \PD^{(n)} ~.

1717: \eeq

1718: 

1719: One can  regard an Apollonian packing $\sP$ 

1720: as a geometric object consisting of an infinite collection of

1721: circles. These circles are described  as the four

1722: orbits 

1723: of a group of M\"{o}bius transformations $G_{\sA}(\sD)$

1724: acting on the circles

1725: $(C_1, C_2, C_3, C_4)$ in

1726: original Descartes configuration $\sD$. The group

1727: \beql{4311}

1728: G_{\sA}(\sD) := \langle \fs_1, \fs_2, \fs_3, \fs_4\rangle, 

1729: \eeq

1730: in

1731: which  $\fs_j$ is the inversion with respect to the circle that

1732: passes through the three intersection points of $\sD$ that do not

1733: include the circle $C_j$, as indicated in \S3.3. One can check

1734: that the circles added at the $n$-th stage of the construction above

1735: correspond to words $\fs_{i_1} \fs_{i_2} \cdots \fs_{i_n}$

1736: of length $n$ in the generators of this group,

1737: in which any two adjacent generators are distinct, i.e. $i_j \ne i_{j+1}$.

1738: 

1739: The key geometric property of Apollonian circle packings

1740: is that they can be viewed as packings of disks having

1741: the circles as boundary. 

1742: To each circle in the Riemann sphere correspond

1743: two  disks having the

1744: circle as boundary. Recall that the ``interior'' of an

1745: oriented circle corresponds to

1746: making  a choice of one of these two disks, 

1747: indicated by a choice of sign of the curvature, when it is

1748: nonzero (and by a choice of normal vector if 

1749: the curvature is zero). 

1750: The disk packing property 

1751: of an Apollonian packing  corresponds to treating its 

1752: Descartes configurations as positively oriented.  

1753: 

1754: %

1755: % theorem 4.1

1756: %

1757: 

1758: \begin{theorem}~\label{th40}

1759: For any  Apollonian packing generated by a

1760: positively oriented Descartes configuration,

1761: the interiors of all circles in the packing

1762: are disjoint.

1763: \end{theorem}

1764: 

1765: \paragraph{Proof.} 

1766: We generate the packing from a single

1767: Descartes configuration $\sD$ chosen to have

1768: positive total orientation, so that the interiors

1769: of its four circles defined by the total orientation

1770: are disjoint. 

1771: Positive total  orientation  is preserved by the action of M\"{o}bius

1772: transformations; see Theorem~\ref{th90a} in Appendix A.

1773: Let 

1774: $$

1775: \sP_{\sD}[m]:=\bigcup_{n=1}^m \sP_{\sD}^{(n)},

1776: $$

1777: where $\sD$ is a positively oriented Descartes configuration. 

1778: We prove by induction on $m$ the following statement: 

1779:  $\sP_{\sD}[m]$ consists of $2(3^m+1)$ circles which form the boundary 

1780: of $2(3^m+1)$ disks and $4\cdot 3^m$ lune areas. The  interiors of the 

1781: disks, as well as the lunes, are mutually disjoint. 

1782: 

1783: The base case $m=0$ is easy. Since any two positively oriented Descartes 

1784: configurations are equivalent under the M\"{o}bius transformations, which

1785: map circles to circles and preserve the (total) orientation, 

1786: (c.f. Theorem 7.2, Appendix A),

1787: we can simply check the Descartes configuration given in Figure 7, which 

1788: has $4$ circles that form the boundary of $4$ disks and $4$ lunes 

1789: with disjoint 

1790: interiors. 

1791: 

1792: Assume the statement holds for $m \geq 0$. We carry out 

1793:  the inductive step for  $m+1$. By the inductive construction above, 

1794: \[

1795: \sP_{\sD}[m+1] = \sP_{\sD}[m] \cup \sP_{\sD}^{(m+1)}.

1796: \]

1797: The circles of $\sP_{\sD}[m]$ remain in $\sP_{\sD}[m+1]$ with their

1798: interiors untouched. In each lune of $\sP_{\sD}[m]$ a new circle is

1799: inscribed, breaking the original lune into 4 parts--one disk and three

1800: smaller lunes, with disjoint interiors. Combining the inductive hypothesis,

1801: we have that  

1802: $\sP_{\sD}[m+1]$ consists of $2(3^m+1)+4\cdot 3^m =2(3^{m+1}+1)$ many 

1803: circles with empty interiors, and the uncovered area are $3\cdot 4\cdot 3^m

1804: =4\cdot 3^{m+1}$ lunes  with disjoint interiors.  

1805: Each new Descartes configuration shares the interior of three circles

1806: with the Descartes configuration  generating it; this implies it has

1807: the same orientation as the preceding one; hence it has positive

1808: total orientation.

1809: This proves the 

1810: statement for $m+1$.  $~~~\bsq$ \\

1811: 

1812: 

1813: We first consider an Apollonian packing as a geometric

1814: object. A {\em geometric Apollonian packing} $\sP$

1815: is the point set

1816: consisting of a countable collection  of circles on the

1817: Riemann sphere $\hat{\cc} = \cc \cup \{\infty\}$

1818: obtained by 

1819: the construction above. It has Hausdorff dimension $1$,

1820: since it is a countable union of sets of Hausdorff dimension $1$

1821: (circles).

1822: Let  $G(\sP)$ denote the group of M\"{o}bius

1823: transformations that leave $\sP$ invariant. This group can be proved

1824: to be a discrete subgroup of the group of all M\"{o}bius transformations,

1825: which acts transitively on the circles in the packing.

1826: The group $G(\sP)$ contains 

1827: $G_{\sA}(\sD)$ above as a subgroup of index $24$,

1828: with cosets given by 

1829: $24$ M\"{o}bius transformations whose effect is to

1830: fix a generating Descartes configuration $\sD$ and permute the

1831: four circles in it.

1832: (These are nontrivial facts, and we do not prove them here.)

1833: 

1834: The geometric Apollonian packing is not a closed set on the

1835: Riemann sphere. We define  $\Lambda(\sP)$ to be its closure on

1836: the Riemann sphere, and call it the {\em  residual set}

1837: of the Apollonian packing. These sets are prototypical examples

1838: of fractal sets, and  

1839: have been much studied; in 1967 Hirst \cite{Hi67}

1840: showed these sets have Hausdorff dimension strictly between

1841: $1$ and $2$. For later use, we

1842: summarize properties of  $\Lambda(\sP)$   in the following result. 

1843: %

1844: %

1845: % theorem 4.2

1846: %

1847: %

1848: 

1849: \begin{theorem}~\label{th41b}

1850: The residual set $\Lambda(\sP)$ of a

1851: geometric  Apollonian packing $\sP$ has the

1852: following properties.

1853: 

1854: (1) $\Lambda(\sP)$  is the complement in

1855: the Riemann sphere of the interiors of

1856: all circles in the packing. Here  ``interior'' is defined

1857: by a positive orientation of a Descartes configuration in

1858: the packing.

1859: 

1860: (2) $\Lambda(\sP)$ has a Hausdorff dimension

1861: which is independent of the packing $\sP$, that satisfies

1862: the bounds 

1863: $$

1864:    1.300197 < \mbox{dim}_{H}(\Lambda(\sP)) < 1.314534.

1865: $$

1866: In particular, $\Lambda(\sP)$   has Lebesgue measure zero.

1867: 

1868: (3) $\Lambda(\sP)$ is the closure of the countable set of

1869: all tangency points of circles in the packing.

1870: 

1871: (4) $\Lambda(\sP)$ is invariant under the action of the

1872: discrete group $G(\sP)$ of M\"{o}bius transformations.

1873: \end{theorem}

1874: 

1875: \paragraph{Remark.} Property (1) is the more usual

1876: definition of residual set of an Apollonian packing.

1877: 

1878: \paragraph{Proof.}

1879: Let $I(\sP)$ be  the open set which is the union of  the

1880: interiors of all circles in the packing. 

1881: It is clear that 

1882: $$

1883: \Lambda(\sP) \subset \Lambda^{*}(\sP) := \hat{\cc} \backslash I(\sP).

1884: $$

1885: The set $\Lambda^{*}(\sP)$ is the usual definition of  the

1886: residual set of an Apollonian packing; we will show

1887: $\Lambda(\sP) =  \Lambda^{*}(\sP).$

1888:   

1889: The result of Hurst \cite{Hi67} that the Hausdorff

1890: dimension of $\Lambda^{*}(\sP)$

1891: is strictly less than $2$  implies

1892: that $\Lambda^{*}(\sP)$

1893:  has Lebesgue measure zero. 

1894: The sharper bounds on the Hausdorff dimension 

1895: stated  here were obtained in 1973 by 

1896: Boyd ~\cite{Bo73b}. A later result of Boyd \cite{Bo82}

1897: suggests that 

1898: $\mbox{dim}_{H}(\Lambda^{*}(\sP)) = 1.3056 \pm .0001.$

1899: Thomas and Dhar\cite{TD94} give a non-rigorous

1900: approximation scheme  suggesting that this Hausdorff dimension

1901: is $1.30568673$ with an error of $1$ in the last digit.

1902: See Falconer \cite[pp. 125--131]{Fa85} for proofs

1903: of $1 < \mbox{dim}_{H}(\Lambda^{*}(\sP)) < 1.432.$

1904: 

1905: Now let $\bz \in \Lambda^{*}(\sP)$. The

1906: interiors of the

1907: circles in the Apollonian packing cover all but a

1908: measure zero area in a neighborhood of  $\bz$.

1909: This requires infinitely many circles,

1910: so their radii must go to zero, since there are only finitely

1911: many circles of radius larger than any positive bound.

1912: Since each of these circles has a tangency point on it,

1913: the point $\bz$ is a limit point of such tangency points.

1914: Since the tangency points are contained in $\sP$,

1915: so we conclude that 

1916: $\Lambda^{*}(\sP) \subset \Lambda(\sP).$

1917: This gives $\Lambda(\sP) = \Lambda^{*}(\sP)$, at

1918: which point properties (1), (2) and (3) follow. 

1919: 

1920: Property (4) follows by

1921: observing that the group invariance of $\sP$ carries over

1922: to its closure, by applying it to any Cauchy

1923: sequence of points in $\sP$. $~~~\bsq$. \\

1924: 

1925: The main viewpoint of this paper is to treat

1926: an Apollonian circle packing $\sP$

1927: as  described ``algebraically''  by

1928: the set of all Descartes configurations it contains,

1929: which we denote $\DD(\sP)$. In  the construction process above, 

1930: except for the four circles in $\sD$, each other  circle

1931: $C$ in the packing corresponds to the unique (unordered) 

1932: Descartes configuration

1933: containing it which occurs at the stage of the construction where $C$ first 

1934: appears. Each (unordered, unoriented) Descartes configuration  appears exactly

1935: once in this correspondence,  except for the base configuration $\sD$

1936: which corresponds to the initial  four circles.

1937: These Descartes configurations can be described in terms of

1938: orbits of a discrete subgroup $\sA$ of  $Aut(Q_{\sD})$

1939: called here the Apollonian group. 

1940: 

1941: 

1942: The Apollonian group $\sA$ is a subgroup of $Aut(Q_{D})$

1943: defined by the action of

1944: the reflection operations on Descartes configurations

1945: described in \S3.3. These correspond to inversions in the circles

1946: passing through the three tangency

1947: points in a Descartes configuration that do not include

1948: one fixed circle in the configuration.

1949: 

1950: \begin{defi}\label{de42}

1951: {\rm 

1952: The {\em Apollonian group} $\sA$ is the 

1953: subgroup

1954: of $Aut(Q_{D})$ defined  by 

1955: \beql{Y318a}

1956: \sA : = \langle \bS_1, \bS_2, \bS_3, \bS_4 \rangle ~,

1957: \eeq

1958: where

1959: \beql{Y319a}

1960: \begin{array}{ccclccc}

1961: \bS_1 & = & \left[

1962: \begin{array}{rccc}

1963: -1 & 2 & 2 & 2 \\

1964: 0 & 1 & 0 & 0 \\ 

1965: 0 & 0 & 1 & 0 \\

1966: 0 & 0 & 0 & 1

1967: \end{array}

1968: \right]~, &~~~~~~ & \bS_2 & = & \left[

1969: \begin{array}{crcc}

1970: 1 & 0 & 0 & 0 \\

1971: 2 & -1 & 2 & 2 \\

1972: 0 & 0 & 0 & 0 \\

1973: 0 & 0 & 0 & 0

1974: \end{array}

1975: \right] ~, \\ [+.2in]

1976: \bS_3 & = & \left[

1977: \begin{array}{ccrc}

1978: 1 & 0 & 0 & 0 \\

1979: 0 & 1 & 0 & 0 \\

1980: 2 & 2 & -1 & 2 \\

1981: 0 & 0 & 0 & 1

1982: \end{array}

1983: \right] ~, &&

1984: \bS_4 & = & \left[

1985: \begin{array}{cccr}

1986: 1 & 0 & 0 & 0 \\

1987: 0 & 1 & 0 & 0 \\

1988: 0 & 0 & 1 & 0 \\

1989: 2 & 2 & 2 & -1

1990: \end{array}

1991: \right] ~.

1992: \end{array}

1993: \eeq

1994: }

1995: \end{defi}

1996: 

1997: 

1998: We now characterize

1999: the set of ordered, oriented Descartes configurations $\DD(\sP)$ 

2000: in an Apollonian

2001: circle packing $\sP$ in terms of this group. We note that

2002: a single unordered, unoriented

2003: Descartes configuration,

2004: corresponds to  exactly 48 ordered,

2005: oriented Descartes

2006: configuration, since there are

2007:  24 choices of ordering of the four circles, and

2008: two choices of orientation.

2009: 

2010: %

2011: % Theorem 4.3

2012: %

2013: 

2014: \begin{theorem}~\label{th41}

2015:  The ordered, oriented Descartes configurations in an Apollonian circle 

2016: packing comprise a  union of 48  orbits of the Apollonian group.

2017: Each of these orbits contains exactly one ordered,

2018: oriented representative of each 

2019: (unordered, unoriented) Descartes configuration in the packing.

2020: \end{theorem}

2021: 

2022: \paragraph{Proof.}

2023: Let $\sD=(C_1, C_2, C_3, C_4)$ be the zeroth stage packing of $\sP$, which is

2024: ordered and positively oriented. It is enough to show that

2025: $\sA(\sD)$ contains exactly one ordered, positively

2026: oriented representative of each

2027: unordered, unoriented Descartes configuration in the packing. Note that

2028: $\bS_1^2=\bS_2^2=\bS_3^2=\bS_4^2=\bI$, 

2029: and for any word $\bS_{i_1}\bS_{i_2}\dots \bS_{i_n}$,

2030: ($i_j \neq i_{j+1}$), the Descartes configuration

2031:  $\sD'=(C_1', C_2', C_3', C_4')$ with ACC coordinates

2032: $\bW_{\sD'} =\bS_{i_1}\bS_{i_2} \dots \bS_{i_n} \bW_\sD$

2033: is a positively oriented Descartes configuration containing a unique circle

2034:   $C_{i_n}'$ in the $n$-th stage partial packing.

2035: Since no two circles at level $n$ touch for $n \geq 1$, by induction

2036: it is easy to see that

2037: \[

2038: \sA=\{ \bS_{i_1} \bS_{i_2} \dots \bS_{i_n} \ | \ n \geq 0, i_j \neq i_{j+1} \},

2039: \]

2040: and

2041: \[

2042: \sA(\sD)=\{ \sD' \ | \  \bW_{\sD'}=\bS_{i_1} \bS_{i_2} \dots 

2043: \bS_{i_n} \bW_\sD ,

2044: \ n \geq 0,

2045: i_j \neq i_{j+1} \}.

2046: \]

2047: Furthermore, for any ordered, positively oriented configuration $\sD''$ in

2048: $\sP$, there is a unique permutation $\sigma$ such that $\sigma(\sD'')=\sD'

2049: \in \sA(\sD)$, where there is a  unique sequence $i_1i_2\dots, i_n$,

2050: $1 \leq i_j \leq 4$ and $i_j \neq i_{j+1}$, such that

2051: $\bW_{\sD'} = \bS_{i_1} \bS_{i_2} \dots \bS_{i_n} \bW_{\sD}$.

2052: This proves that all the ordered, positively oriented Descartes configurations

2053: in an Apollonian circle packing comprise a union of 24 orbits of the Apollonian

2054: group. The theorem follows by counting both the positively and negatively

2055: oriented Descartes  configurations.

2056:  $~~~\bsq$ \\

2057: 

2058: Theorem~\ref{th41}  shows  that the orbit under

2059: the Apollonian group of a single (ordered, oriented)

2060: Descartes configuration completely describes an Apollonian packing.

2061: 

2062: 

2063: %*****************************************************************************

2064: %

2065: % SECTION 5 

2066: %

2067: %*****************************************************************************

2068: %

2069: \section{Dual Apollonian Group}

2070: 

2071: 

2072: As explained in \S3, the operation of 

2073: inversion in the each of 

2074: the individual circles of a Descartes configuration

2075: is described by an integral matrix in $Aut(Q_{D})$.

2076: This leads us to the following definition.

2077: 

2078: \begin{defi}\label{de43}

2079: {\rm The {\em dual Apollonian group} $\sA^{\perp}$

2080: is the subgroup of $Aut(Q_{D})$ generated by the

2081: matrices 

2082: \beql{S413}

2083: \begin{array}{rllcrll}

2084: \bS_1^\perp  & = & \left[

2085: \begin{array}{rccc}

2086: -1 & 0 & 0 & 0 \\

2087: 2 & 1 & 0 & 0 \\

2088: 2 & 0 & 1 & 0 \\

2089: 2 & 0 & 0 & 1

2090: \end{array}

2091: \right] ~, & \qquad & \bS_2^\perp & = & \left[

2092: \begin{array}{crcc}

2093: 1 & 2 & 0 & 0 \\

2094: 0 & -1 & 0 & 0 \\

2095: 0 & 2 & 1 & 0 \\

2096: 0 & 2 & 0 & 1

2097: \end{array}

2098: \right] ~, \\ [+.3in]

2099: \bS_3^\perp & = & \left[

2100: \begin{array}{ccrc}

2101: 1 & 0 & 2 & 0 \\

2102: 0 & 1 & 2 & 0 \\

2103: 0 & 0 & -1 & 0 \\

2104: 0 & 0 & 2 & 1

2105: \end{array}

2106: \right] ~, && \bS_4^\perp & = & \left[

2107: \begin{array}{cccr}

2108: 1 & 0 & 0 & 2 \\

2109: 0 & 1 & 0 & 2 \\

2110: 0 & 0 & 1 & 2 \\

2111: 0 & 0 & 0 & -1

2112: \end{array}

2113: \right] ~.

2114: \end{array}

2115: \eeq

2116: }

2117: We note that $\bS_i^{\perp}= \bS_i^{T}$, the transpose of $\bS_i$.

2118: \end{defi}

2119: 

2120: 

2121: By analogy with

2122: Theorem~\ref{th41} we  might

2123: think of  an orbit of the dual Apollonian group acting on

2124: a single oriented Descartes configuration $\sD$,

2125: as a  ``dual Apollonian circle packing''.

2126: It can be viewed algebraically as a set of Descartes configurations

2127: or geometrically as a collection of circles in the plane. 

2128: In the algebraic viewpoint the orbit is a discrete set of points in

2129: the parameter space $\sM_{\dd}$ so is a discrete object. 

2130: From  the  geometrical viewpoint,

2131: as a collection of circles, which we denote $\sP_{\sD}^{\perp}$,

2132: it has the 

2133: following weak ``packing''  property.

2134: %

2135: % Theorem 5.1

2136: %

2137: 

2138: \begin{theorem}~\label{th41a} 

2139: No  two circles in a dual Apollonian circle packing

2140: cross each other. That is, two circles  in distinct

2141: Descartes configurations in a dual Apollonian circle packing

2142: $\sP_{\sD}^{\perp}$

2143: either coincide, or are tangent to each other, or

2144: are disjoint.

2145: \end{theorem}

2146: 

2147: We do not give a proof of this theorem here, as it 

2148: follows from a similar result proved for the

2149: super-Apollonian packing in \S3 of part II.

2150: It also is a special case of an $n$-dimensional generalization

2151: proved  in \S4.2 of part III.

2152: 

2153: Figure~\ref{dualpack}

2154: pictures the  circles in a dual packing $\sP^{\perp}$. 

2155: These are circles in Descartes configurations dual

2156: to those in the Apollonian packing in Figure \ref{std}.

2157: The fractal-like part of this figure is the closure of

2158: the set of points at which circles  in the dual packing touch.

2159: It can be shown that  infinitely many circles touch at each such tangency

2160: point. Below we shall show that this fractal set coincides with the

2161: limit set of the Apollonian packing generated by the

2162: dual Descartes configuration $\bD(\sD_0)$. 

2163: %

2164: % figure 7a

2165: %

2166: 

2167: \begin{figure}[htbp]

2168: \centerline{\epsfxsize=5.0in \epsfbox{dualpack.ps}}

2169: \caption{A dual Apollonian packing}\label{dualpack}

2170: \end{figure}

2171: 

2172: The  dual Apollonian group has a

2173: simple relation  to the Apollonian group.

2174: 

2175: %

2176: % Theorem 5.2

2177: %

2178: 

2179: \begin{theorem}~\label{th42}

2180: The dual Apollonian group $\sA^{\perp}$ is conjugate to the Apollonian group

2181: $\sA$ using  the duality operator $\bD \in Aut(Q_D)$. This holds

2182: at the level of generators, with 

2183: $$

2184: \bD^T \bS_i \bD = \bS_i^{\perp} ~~\mbox{for}~ 1 \le i \le 4.

2185: $$

2186: where $\bD= \bD^{-1} = \bD^T$.

2187: \end{theorem}

2188: 

2189: \paragraph{Proof.} This is a straightforward computation.  $~~~\bsq$

2190: 

2191: %

2192: % Theorem 5.3

2193: %

2194: 

2195: \begin{theorem}~\label{th43}

2196: Given a dual Apollonian packing 

2197: $\sP_{\sD_0}^{\perp}$ generated by Descartes configuration $\sD_0$, 

2198: let $\Sigma$

2199: be the set of all intersection points

2200: of the circles in $\sA^{\perp}(\sD_0)$. 

2201: Then the  closure $\Lambda := \overline{\Sigma}$

2202:  is equal to the residual set $ \Lambda(\sP_{\bD(\sD_0)})$ 

2203: of the Apollonian packing $\sP_{\bD(\sD_0)}$ generated by

2204: the dual Descartes configuration

2205: $ \bD(\sD_0)$.

2206: \end{theorem}

2207: 

2208: \paragraph{Proof.}

2209: Theorem~\ref{th41b}(3) states

2210: that   the residual set $\Lambda(\sP)$ of a

2211: geometric  Apollonian

2212: packing $\sP$ is the closure of the 

2213: set of tangency points $\Sigma'$ in $\sP$.

2214: Thus it will suffice to show that the set of

2215: tangency points $\Sigma$ of the circles in 

2216: the dual packing $\sP_{\sD_0}^{\perp}$

2217: coincides with the set of tangency points

2218: $\Sigma'$ of circles in the Apollonian packing

2219: $\sP_{\bD (\sD_0)}$.

2220: Each intersection point of two circles in the dual

2221: Apollonian packing $\sP_{\sD_0}^{\perp}$

2222: is an intersection point of circles in some Descartes configuration

2223: $\sD$ of $\sA^{\perp}(\sD_0)$. (This follows from the recursive

2224: construction of the dual packing.)

2225: Each intersection point or circles in $\sD$ is 

2226: an intersection point of circles in the dual Descartes 

2227: configuration $\bD(\sD)$, which is the configuration

2228: having augmented curvature-center coordinates

2229: $\bD \bW_{\sD}$. Now $\bD(\sD)$

2230: belongs to the Apollonian packing

2231: $\sA(\bD (\sD_0))$ because

2232: $$

2233: \bD \bW_{\sD} = \bD (\bS_{i_1}^{\perp} \cdots \bS_{i_m}^{\perp} \bW_{\sD_0})=

2234: \bS_{i_1} \cdots \bS_{i_m} (\bD \bW_{\sD_0}),

2235: $$

2236: using Theorem~\ref{th42}. Thus the set $\Sigma$ 

2237: of intersection points

2238: of circles  in the dual Apollonian packing 

2239: $\sP_{\sD_0}^{\perp}$ is contained in 

2240: the set $\Sigma'$ of intersection points of circles in 

2241: the Apollonian packing $\sP_{\bD(\sD_0)}$.

2242: The argument reverses to

2243: show the converse is also true: each

2244: intersection point of circles in the Apollonian

2245: packing $\sP_{\bD (\sD_0)}$ is an intersection point of circles

2246: in the dual Apollonian packing  $\sP_{\sD_0}^{\perp}$.

2247: Thus $\Sigma = \Sigma'$. $~~~\bsq$

2248: 

2249: 

2250: 

2251: 

2252: We have the following dichotomy between 

2253: the geometric and algebraic views of Apollonian

2254: packings and dual Apollonian packings.

2255: Viewed geometrically as collections of circles,

2256: a dual Apollonian packing $\sP^{\perp}$ 

2257: appears quite  different from an Apollonian

2258: packing, as the circles are nested to an infinite depth. 

2259: However when viewed  algebraically as a collection of 

2260: ordered, oriented Descartes

2261: configurations $\DD(\sP^{\perp})$, it is a discrete set

2262: given by the orbit of a group $\sA^{\perp}$ conjugate

2263: to the Apollonian group $\sS$, and so is a similar object.

2264: 

2265: %*****************************************************************************

2266: %

2267: % SECTION 6. 

2268: %

2269: %*****************************************************************************

2270: %

2271: 

2272: \section{Super-Apollonian group}

2273: \setcounter{equation}{0}

2274: 

2275: 

2276: The Apollonian group $\sA$ and dual Apollonian group $\sA^{\perp}$

2277: both consist of integer automorphs in $Aut(Q_{D})$. We 

2278: obtain a larger integral group by combining the

2279: two groups, as follows.

2280: 

2281: \begin{defi}\label{de44}

2282: {\rm The {\em super-Apollonian group} $\sA^{S}$ is the

2283: subgroup of $Aut(Q_{D})$ generated by the Apollonian

2284: group $\sA$ and $\sA^{\perp}$ together. We have

2285: $$

2286: \sA^{S} = \langle \bS_1, \bS_2, \bS_3, \bS_4, \bS_1^{\perp},  \bS_2^{\perp},

2287: \bS_3^{\perp},  \bS_4^{\perp} \rangle.

2288: $$

2289: }

2290: \end{defi}

2291: 

2292: The super-Apollonian group  $\sA^{S}$ is a discrete subgroup

2293: of $Aut(Q_D)$ as a consequence of the fact that all its members

2294: are integral matrices. 

2295: 

2296: 

2297: \subsection{Presentation of super-Apollonian group}

2298: 

2299: 

2300: We now determine a presentation of

2301:  the  super-Apollonian group, showing that it  

2302: is a hyperbolic Coxeter group.

2303: This result implies that both the 

2304: Apollonian group and dual Apollonian group 

2305: are finitely presented, and are hyperbolic 

2306: Coxeter groups.

2307: 

2308: \begin{theorem}~\label{Sth47}

2309: The  super-Apollonian group

2310: $$

2311: \sA^{S} = \langle \bS_1,~\bS_2,~\bS_3,~\bS_4,

2312: ~\bS_1^{\perp},~\bS_2^{\perp},~\bS_3^{\perp},~\bS_4^{\perp} \rangle

2313: $$

2314: is a hyperbolic Coxeter group

2315: whose complete set of Coxeter relations are

2316: \beql{S422b}

2317: \bS_i^2 = ~(\bS_i^\perp)^2 = ~\bI \quad\mbox{for}\quad 1 \leq i \leq 4.

2318: \eeq

2319: \beql{S423}

2320: (\bS_i \bS_j^\perp)^2~= ~(\bS_j^\perp \bS_i)^2 = ~\bI \quad\mbox{if}\quad i \neq j.

2321: \eeq

2322: \end{theorem}

2323: 

2324: \paragraph{Proof.}

2325: Recall that $\bS_i^\perp = \bS_i^T.$

2326: The group $\sA^{S}$ satisfies the relations

2327: $\bS_i^2 = (\bS_i^T)^2 =\bI$ for $1 \le i \le 4$ and it satisfies

2328: \beql{U429}

2329: \bS_i \bS_j^T = \bS_j^T \bS_i \quad\mbox{for}\quad

2330: i \neq j ~,

2331: \eeq

2332: which is equivalent to

2333: \eqn{S423}.

2334: 

2335: Write words $\bU$ in $\sA^{S}$ as $\bU = \bU_n \bU_{n-1} \cdots \bU_1$,

2336:  in which each $\bU_i$ is a generator $\bS_j$ or $\bS_j^T$, with the empty

2337: word (n = 0) being the identity element.

2338: We do not need inverses since each generator is its own inverse.

2339: A word $\bU$ is in {\em normal form} if it has the two properties:

2340: \begin{itemize}

2341: \item[(i)]

2342: $\bU_{k} \neq \bU_{k - 1} $ for $2 \le k \le n$.

2343: \item[(ii)]

2344: If $\bU_k = \bS_j^T$ for some $j$, then 

2345: $\bU_{k - 1} \neq \bS_i$ for all $i \neq j$.

2346: \end{itemize}

2347: We can reduce any word to a word in normal form, using the relations

2348: to move any symbols $\bS_j^T$ as far to the right in the word

2349: as possible.  We cancel any adjacent identical symbols to make (i) hold.

2350: Then move the rightmost $\bS_j^T$ as far to the right as possible,

2351: as allowed by \eqn{U429}.

2352: Repeat the same with the next rightmost symbol $\bS_j^T$.

2353: If in the process any adjacent symbols $(\bS_j^T)^2$ occur, cancel them.

2354: This process must terminate in a normal form word or the empty word.

2355: 

2356: The theorem is equivalent to showing that no normal form word with

2357: $n \ge 1$ is the identity element in $\sA^{S}$.

2358: This holds because the

2359:  reduction to normal form used only the Coxeter relations,

2360: so if $\sA^{S}$ satisfies an additional nontrivial relation,

2361: there would exist some nontrivial normal form word that is the identity.

2362: 

2363: We proceed by induction on the length $n$ of a word in normal form

2364: $\bU =\bU_n \bU_{n-1} \cdots \bU_1$.

2365: Note that any suffix $\bU_{j-1} \bU_j \cdots \bU_1$ 

2366: of $\bU$ is also in normal form.

2367: For each $n \ge 1$, let $\sX_n$ denote

2368: the set of  all  normal form words of length $n$.

2369: (By a simple enumeration we can show that  the number of normal form words 

2370: of length $n$ is $9 \cdot 5^{n-1}-1$. ) 

2371: We measure the {\em size} of a word $\bU$ 

2372: (viewed as a $4 \times 4$ matrix) as

2373: \beql{U430}

2374: f(\bU) := \bo^T \bU \bo = \sum_{i=1}^4 \sum_{j=1}^4 \bU_{ij} ~.

2375: \eeq

2376: Thus $f(\bI) = 4.$ For $n=1$ we have

2377: \beql{U431}

2378: f(\bU) = f(\bU_1) = 8

2379: \eeq

2380: in all cases.

2381: For $n \ge 2$ if $\bU \in \sX_n$ then

2382: $\bU' = \bU_{n-1} \bU_{n-2} \cdots \bU_1 \in \sX_{n-1}$, and we will  prove

2383: \beql{U432}

2384: f(\bU) > f(\bU' ) ~.

2385: \eeq

2386: If so, then

2387: $$f(\bU) \ge f(\bU_1) \ge 8, $$

2388: hence $\bU \neq \bI$, which will complete the proof.

2389: 

2390: We let

2391: \beql{U433}

2392: \br (\bU) := (r_1 ( \bU) , r_2 (\bU) , r_3 (\bU) , r_4 (\bU))^T = \bU \bo \,,

2393: \eeq

2394: be the vector of row sums of $\bU$. Now

2395: \begin{eqnarray*}

2396: f(\bS_1 \bU') & = & \bo^T

2397: \bS_1 \br (\bU') = (-1,~ 3,~ 3,~ 3) \br (\bU') = -4 r_1 (\bU') + 3 f(\bU' ) \,, \\

2398: f(\bS_1^T \bU') & = & \bo^T \bS_1^T \br (\bU') = (5,~1,~1,~1) \br (\bU)

2399: = 4r_1 (\bU') + f(\bU'),

2400: \end{eqnarray*}

2401: with similar formulas in the other cases.

2402: To prove \eqn{U432} it is therefore sufficient to prove the

2403: following two assertions,

2404: for $1 \le h \le 4$.

2405: \begin{itemize}

2406: \item[(1)]

2407: If $\bU_n = \bS_h$, so that $\bU_{n-1} \neq \bS_h$, then 

2408: $2r_h (\bU') < f(\bU')$.

2409: \item[(2)]

2410: If $\bU_n = \bS_h^T$, so that $U_{n-1} \neq \bS_h^T$ and 

2411: $U_{n - 1} \neq \bS_i$

2412: for $i \neq h$,

2413: then $r_h (\bU') > 0$.

2414: \end{itemize}

2415: 

2416: \noindent

2417: 

2418: Instead of proving (1), (2), we

2419:  will prove by induction on $n$ the following three assertions for all

2420: $\bU \in \sX_n$. Here  $(h,i,j,k)$ always denotes some permutation of

2421: $(1,2,3,4)$ in what follows.

2422: 

2423: \begin{itemize}

2424: \item[(i)]

2425: For $1 \le i,j \le 4$, with $i \neq j$,

2426: $$r_i (\bU) + r_j (\bU) > 0 ~.$$

2427: \item[(ii)]

2428: If $\bU_n = \bS_h$, then

2429: \begin{eqnarray*}

2430: r_h (\bU) & > & 0, \\

2431: r_i (\bU) & < & r_h (\bU) + r_j (\bU) + r_k (\bU), \\

2432: r_j (\bU) & < & r_h (\bU) + r_i (\bU) + r_k (\bU), \\

2433: r_k (\bU) & < & r_h (\bU) + r_i (\bU) + r_j (\bU)~.

2434: \end{eqnarray*}

2435: 

2436: \item[(iii)]

2437: If $\bU_n = \bS_h^T$ then

2438: \begin{eqnarray*}

2439: r_h (\bU) & < & 0, \\

2440: r_i (\bU) & < & r_h (\bU) + r_j (\bU) + r_k (\bU), \\

2441: r_j (\bU) & < & r_h (\bU) + r_i (\bU) + r_k (\bU), \\

2442: r_k (\bU) & < & r_h (\bU) + r_i (\bU) + r_j (\bU) ~.

2443: \end{eqnarray*}

2444: \end{itemize}

2445: 

2446: \noindent

2447: Note that (i) implies that at most one of the row sums of $\bU$ 

2448: can be negative.

2449: If proved, (i)--(iii) imply (1), (2), which themselves imply \eqn{U432},

2450: completing the proof of the Theorem.

2451: 

2452: The induction hypothesis (i)--(iii) holds for $n=1$, since

2453: $r(\bS_h)$ is a permutation of $(5,1,1,1)^T$ and $r(\bS_h^T)$

2454: is a permutation of $(-1,3,3,3)^T$.

2455: Suppose it is true for $n$.

2456: For the induction step, write $\bU \in \sX_{n+1}$ as

2457: $\bU = \bU_{n+1} \bU'$ with $\bU' \in \sX_n$ 

2458: and abbreviate $r'_i = r_i (\bU')$.

2459: 

2460: \paragraph{Case 1.}

2461: {\em $U_{n+1} = \bS_h$ for some $h$.}

2462: 

2463: \noindent

2464: We must verify (i) and (ii) for $\bU$.

2465: We have

2466: $$

2467: \begin{array}{lllll@{}l}

2468: r_h (\bU) & = & r_h (\bS_h \bU') & = & - & r'_h + 2r'_i + 2r'_j + 2r'_k, \\

2469: r_i (\bU) & = & r_i (\bS_h \bU') & = & r'_i, \\

2470: r_j (\bU) & = & r_j (\bS_h \bU') & = & r'_j, \\

2471: r_k (\bU) & = & r_k (\bS_h \bU') & = & r'_k ~.

2472: \end{array}

2473: $$

2474: To verify (i), all cases not involving $r_h (\bU)$ follow from the induction

2475: hypothesis.

2476: To show $r_h (\bU) + r_i (\bU) > 0$ note that

2477: \beql{U433a}

2478: r_h (\bU) + r_i (\bU) = - r'_h + 3r'_i + 2r'_j + 2r'_k

2479: ~.

2480: \eeq

2481: Now $\bU_{n} \neq \bS_h$ since $\bU \in \sX_{n+1}$, and for all cases

2482: except $\bU_{n} = \bS_h^T$ the induction hypotheses (ii) and (iii) give

2483: $r'_h < r'_i + r'_j + r'_k$, whence

2484: $$r_h (\bU) + r_i (\bU) \ge 2r'_i + r'_j + r'_k =

2485: (r'_i + r'_j ) + (r'_i + r'_k) > 0 \,,

2486: $$

2487: using induction hypothesis (i).

2488: If $\bU_{n} = \bS_h^T$, then by hypothesis (iii) $r'_h < 0$,

2489: hence $r'_i, r'_j , r'_k > 0$ by

2490: hypothesis (i), so all terms on the right side of \eqn{U433a} are positive,

2491: so $r_h (\bU) + r_i (\bU) > 0$.

2492: The cases

2493: $r_h (\bU) + r_j (\bU) > 0$ and $r_h (\bU) + r_k (\bU) > 0$ follow similarly.

2494: To verify (ii) for $\bU$ it suffices to prove

2495: \begin{eqnarray*}

2496: && 0 < - r'_h + 2r'_i + 2r'_j + 2r'_k, \\

2497: && r'_i < -r'_h + 2r'_i + 3r'_j + 3r'_k, \\

2498: && r'_j < - r'_h + 3r'_i + 2r'_j + 3r'_k, \\

2499: && r'_k < - r'_h + 3r'_i + 3r'_j + 2r'_k,

2500: \end{eqnarray*}

2501: which is equivalent to

2502: \begin{eqnarray}\label{U434}

2503: r'_h & < & 2r'_i + 2r'_j + 2r'_k, \nonumber \\

2504: r'_h & < & r'_i + 3r'_j + 3r'_k, \nonumber \\

2505: r'_h & < & 3r'_i + r'_j + 3r'_k,  \\

2506: r'_k & < & 3r'_i + 3r_j + r'_k ~. \nonumber

2507: \end{eqnarray}

2508: There are three cases, according as

2509: $\bU_n = \bS_i$, $\bU_n = \bS_i^T$ with $i \neq h$ and $\bU_n = \bS_h^T$.

2510: In the first two of these,

2511: the induction hypotheses give

2512: $$r'_h < r'_i + r'_j + r'_k,$$

2513: which with induction hypothesis (i) immediately yields \eqn{U434}.

2514: If $\bU_n = \bS_h^T$ then by (iii), $r'_h < 0$, whence

2515: $r'_i , r'_j , r'_k > 0$ by (i) so \eqn{U434} is immediate.

2516: This finishes Case 1.

2517: 

2518: \paragraph{Case 2.}

2519: {\em $\bU_{n+1} = \bS_h^T$ for some $h$.}

2520: 

2521: \noindent

2522: We must verify (i) and (iii).

2523: We have

2524: $$

2525: \begin{array}{lllll}

2526: r_h (\bU) & = & r_h (\bS_h^T \bU' ) & = & -r'_h, \\

2527: r_i (\bU) & = & r_i (\bS_h^T \bU') & = & 2r'_h + r'_i, \\

2528: r_j (\bU) & = & r_j (\bS_h^T \bU') & = & 2r'_h + r'_j, \\

2529: r_k (\bU) & = & r_k (\bS_h^T \bU') & = & 2r'_h + r'_j,

2530: \end{array}

2531: $$

2532: To prove (i), note first that

2533: $$r_h (\bU) + r_i (\bU) = r'_h + r'_i > 0,$$

2534: using (i) for $\bU'$, and similar inequalities hold

2535:  for other cases involving $r_h (\bU)$.

2536: The remaining cases are all of the form

2537: $r_i (\bU) + r_j (\bU) > 0$

2538: with $i,j \neq h$, and the proofs for each are similar.

2539: We have

2540: $$

2541: r_i (\bU) + r_j (\bU) = 4r'_h + r'_i + r'_j =

2542: 2r'_h + (r'_h + r'_i) + (r'_h + r'_j ) > 2r'_h \,,

2543: $$

2544: using induction hypothesis (i).

2545: We now show $r'_h > 0$ holds in all cases.

2546: Since $\bU \in \sX_{n+1}$, we have

2547: $\bU_{n} \neq \bS_h^T$ and $\bU_{n} \neq \bS_i$ with $i \neq h$.

2548: If $\bU_{n} = \bS_j^T$ for some $j \neq h$ then hypothesis (iii) for

2549: $\bU'$ says that $r'_j < 0$, whence we must have $r'_h > 0$.

2550: If $\bU_{n} = \bS_h$ then hypothesis (ii) for $\bU'$ says $r'_h > 0$.

2551: Thus $r'_h > 0$ in all cases, and (i) holds for $\bU$.

2552: To prove (ii) we must prove

2553: $$

2554: \begin{array}{l}

2555: r'_h < 0, \\

2556: 2r'_h + r'_i < 3r'_h + r'_j + r'_k, \\

2557: 2r'_h + r'_j < 3r'_h + r'_i + r'_k, \\

2558: 2r'_h + r'_k < 3r'_h + r'_i + r'_j,

2559: \end{array}

2560: $$

2561: which is equivalent to

2562: \beql{U435}

2563: \begin{array}{l}

2564: r'_h > 0, \\

2565: r'_i < r'_h + r'_j + r'_k, \\

2566: r'_j < r'_h + r'_i + r'_k, \\

2567: r'_k < r'_h + r'_i + r'_j ~.

2568: \end{array}

2569: \eeq

2570: We verified $r'_h > 0$ already.

2571: For the remainder there are two cases, according as

2572: $\bU_n = \bS_h$ or $\bU_n = \bS_i^T$ for some $i \neq h$.

2573: If $\bU_n = \bS_h$ the three remaining inequalities in \eqn{U435}

2574: follow from inductive hypothesis (ii) for $\bU'$.

2575: If $\bU_n = \bS_i^T$ the inductive hypothesis (iii) gives the last two

2576: inequalities in \eqn{U435}, and also that $r'_i < 0$.

2577: By hypothesis (i) for $\bU'$ this implies $r'_j , r'_h , r'_k > 0$ hence

2578: $r'_i < 0 < r'_h + r'_j + r'_k$ which verifies \eqn{U435} in this case,

2579: and finishes Case 2.

2580: 

2581: The induction is complete.~~~$\bsq$

2582: %

2583: %

2584: % sect. 6.2

2585: %

2586: 

2587: \subsection{Apollonian super-packings}

2588: 

2589: We now define an {\em Apollonian super-packing}

2590: to be a set of (ordered, oriented) Descartes configurations forming   

2591: an orbit of the super-Apollonian group $\sA^{S}$,

2592: acting on a single such configuration. 

2593: 

2594: The study of Apollonian super-packings will form the subject matter

2595: of part II. We call the set of circles in all the

2596: Descartes configurations in such a super-packing

2597: a {\em geometric Apollonian super-packing}. 

2598: These circles comprise four orbits of a group of

2599: M\"{o}bius transformations

2600: $$

2601: \sG_{\sA^{S}}(\sD) = \langle \fs_1, \fs_2, \fs_3, \bs_4, \bs_1^{\perp},

2602: \bs_2^{\perp}, \bs_3^{\perp}, bs_4^{\perp}\rangle 

2603: $$

2604: where $\fs_i$ and $\fs_i^{\perp}$

2605: were defined in \S3.3. In terms of the space

2606: $\sM_{\DD}$ of Descartes configurations this group action

2607: is given by 

2608: $\tilde{\sG}_{\sA^{S}}(\sD) \subset Aut(Q_{W})$ 

2609: defined by 

2610: $$

2611: \tilde{\sG}_{\sA^{S}}(\sD) := \bW_{\sD}^{-1}  \sA^{S} \bW_{\sD},

2612: $$

2613: using the isomorphism $Aut(Q_{W}) = \mbox{M\"{o}b}(2) \times \{ \bI, -\bI\}$

2614: given in Theorem~\ref{th91} of Appendix A. 

2615: 

2616: 

2617: In part II we  will show that 

2618: the individual circles in a geometric super-packing

2619: form a ``packing'' in

2620: the weak sense that no two circles cross each

2621: other, although circles can be nested.

2622: This is a remarkable geometric fact, because 

2623: we will also show in part II that a 

2624: strongly integral Apollonian super-packing necessarily contains a copy of

2625: every integral Apollonian circle

2626: packing, with all these copies (essentially)

2627:  contained inside the square $0 \le x \le 2, 0 \le y \le 2$.

2628: All these Apollonian packings

2629: fit together in such a way that all the circles in these packings 

2630: manage not to cross each other.

2631: 

2632: %*****************************************************************************

2633: %

2634: % SECTION 9. Appendix A. M\"{o}bius Action

2635: %

2636: %*****************************************************************************

2637: %

2638: \newpage

2639: \section{Appendix A.  M\"{o}bius Group Action}

2640: \setcounter{equation}{0}

2641: The {\em (general) M\"{o}bius group} 

2642:  $\mbox{M\"{o}b}(2)$

2643: is the group of M\"{o}bius transformations, 

2644: allowing reflections

2645: (including complex conjugation).  This group is denoted $GM(\hat{\rr}^2)$ in 

2646: Beardon~\cite[p. 23]{Be83},

2647: and is also known as the {\em conformal group}. 

2648: The group

2649: $\mbox{M\"{o}b}(2)_{+}  \simeq PSL(2, \cc) = SL(2, \CC)/\{\pm \bI \}$

2650: consists of the orientation-preserving maps of 

2651: $\hat{\cc}= \rr^2 \cup \{\infty\}$.

2652: The group $\mbox{M\"{o}b}(2)$

2653: is  a six-dimensional real Lie group which has two connected

2654: components corresponding to orientation-preserving

2655: \footnote{The terminology ``orientation'' here 

2656: refers to the invariant $\det(\bM)$.} and

2657: orientation-reversing transformations.

2658: It can be written as a semi-direct product

2659: $$

2660: \mbox{M\"{o}b}(2) \cong \mbox{M\"{o}b}(2)_{+} \rtimes \{ 1, \fc\},

2661: $$

2662: in which $\fc$ denotes complex conjugation.

2663: M\"{o}bius transformations

2664: take circles to circles (or straight lines)

2665: and preserve angles.

2666: Thus they take ordered Descartes configurations to ordered 

2667: Descartes configurations. Concerning orientation, we show

2668: that although  M\"obius transformations can reverse orientation of a

2669: single circle, they preserve (total) orientation of 

2670: oriented Descartes configurations.

2671: Recall that the {\em total orientation} 

2672: of an oriented Descartes configuration is the sign of the sum of the

2673: (signed) curvatures of the circles in it.

2674: 

2675: 

2676: \begin{theorem}~\label{th90a}

2677: M\"obius transformations preserve the total orientation of

2678: oriented Descartes configurations.

2679: \end{theorem}

2680: 

2681: \pf

2682: Positively oriented Descartes configurations are characterized

2683: by the four circles having disjoint ``interiors'', specified  by

2684: the positive orientation. 

2685: M\"obius transformations take Descartes configurations

2686: to Descartes configurations, and preserve the ``disjoint

2687: interior'' property, hence preserve positive orientation, as

2688: given by normal vectors to the circle. The result for

2689: negatively oriented Descartes configurations holds since

2690: all normal vectors are reversed from the positively oriented  case.

2691: ~~~$\bsq$ \\

2692: 

2693: 

2694: We now consider a group $GM^{*}(2)$

2695: with four connected components,

2696: which we will term 

2697: the  {\em extended General M\"{o}bius group},

2698: defined by 

2699: $GM^{*}(2) := \mbox{M\"{o}b}(2) \times \{ \bI, - \bI\}$.

2700: Here $\{ \bI, -\bI\}$ are in the center of this group,

2701: and we  write elements of $GM^{*}(2)$ as $\pm \fg$,

2702: in which  $\fg \in \mbox{M\"{o}b}(2)$, 

2703: and the sign indicates

2704: which of  $\pm \bI$ occurs.

2705: We have 

2706: $$

2707: GM^{*}(2) \cong SL(2, \CC) \rtimes \{ \bI, \fc \}.

2708: $$

2709: 

2710: %

2711: % Table A-1

2712: %

2713: 

2714: \begin{table}[t]

2715:  \begin{tabular}[t]{|c|c|}

2716:  \hline

2717:   \begin{diagram}  

2718:    GM^{*}(2)=  M\ddot{o}b(2)\times \{-\bI, \bI\}    & \rTo^\sim & Aut(Q_W)  & 

2719:          \rTo^\sim  &   {O(3,1)}  \\

2720:    \dTo^\pi \uInto &   &  \dTo^\pi \uInto &    & \dTo^\pi \uInto  \\

2721:    M\ddot{o}b(2)=PSL(2, \CC)\rtimes \{I, \fc\} & \rTo^\sim & 

2722: Aut(Q_W)^\uparrow 

2723:          &  \rTo^\sim & O(3,1)^\uparrow  \\ 

2724:    \uInto &  &  \uInto& & \uInto  \\

2725:    {M\ddot{o}b(2)_+=PSL(2,\CC) }      & \rTo^{\sim} 

2726:          & Aut(Q_W)^\uparrow_+ &  \rTo^\sim  &  O(3,1)^\uparrow_+  \\ 

2727:   \end{diagram}

2728:  &  

2729:   \begin{diagram} 

2730:   \text{Lorentz group} \\ 

2731:    {} \\

2732:   {\text{orthochronous} \atop  \text{Lorentz group}} \\ 

2733:    {}\\

2734:   {\text{proper orthochronous}  \atop \text{Lorentz  group}} \\

2735:  \end{diagram} 

2736:  \\

2737:  \hline

2738: \end{tabular}

2739: \caption{Group Isomorphisms}~\label{table1}

2740: \end{table}

2741: 

2742: The main object of this Appendix is to  define 

2743: an action of $GM^{*}(2)$ on the right on the parameter space $\sM_{\dd}$,

2744: given in the next theorem. This amounts to

2745: finding an explicit isomorphism between $GM^{*}(2)$

2746: and $Aut(Q_{W})$, which appears as the  horizontal

2747: arrow  on the left in the top row in Table ~\ref{table1}.

2748: This map when restricted to the smaller groups

2749: M\"{o}b(2) and M\"{o}b$(2)_{+}$ give the other

2750: two horizontal isomorphisms on the left side of

2751: the table. 

2752: Table A-1 also indicates isomorphisms on its right side  

2753: to the

2754: Lorentz group $O(3,1)$ and 

2755: corresponding  subgroups, which we defer discussing 

2756: until  after the following result.

2757: 

2758: %

2759: %

2760: %  Theorem A-2

2761: %

2762: 

2763: 

2764: \begin{theorem}\label{th91}

2765: Let $GM^{*}(2) := \mbox{M\"{o}b}(2) \times \{ \bI, - \bI\}$.

2766: There is a unique 

2767: isomorphism $\pi: GM^{*}(2) \to Aut(Q_W)$,

2768: with image elements $ \bV_{\pm \fg}:=\pi(\pm \fg)$,

2769: such that the following hold.

2770:  

2771: (i) For  $\fg \in  \mbox{ M\"{o}b}(2)$

2772: the augmented curvature-center coordinates

2773: for each ordered, oriented Descartes configuration $\sD$ 

2774: satisfy 

2775: \beql{N9.2}

2776: \bW_{\fg(\sD)} = \bW_{\sD} \bV_{\fg}^{-1}.

2777: \eeq

2778: 

2779: 

2780: (ii) The action of $-\bI$ on augmented

2781: curvature-center coordinates is

2782: \beql{N9.2b}

2783: \bW_{-\sD} = \bW_{\sD}\bV_{-\bI}^{-1}= -\bW_{\sD}.

2784: \eeq 

2785: \end{theorem}

2786: 

2787: \paragraph{Proof.}

2788: We compute the action of 

2789: $\mbox{M\"{o}b}(2)$

2790: acting on

2791: augmented curvature-center coordinates.

2792: Let $(\bar{b}, b, w_1, w_2)= (\frac{x_1^2+y_1^2-r^2}{r}, 

2793: \frac{1}{r}, \frac{x_1}{r}, \frac{y_1}{r})$

2794: be the augmented curvature-center coordinates of the circle

2795: $$ (x - x_1)^2 + (y - y_1)^2 = r^2. $$

2796: This circle can be recovered from these coordinates via

2797: \beql{281}

2798: (bx - w_1)^2 + (by - w_2)^2 = 1,

2799: \eeq

2800: and the orientation of the circle (inside versus outside) is

2801: determined by the sign of $b$. An oriented ``circle at infinity''

2802: is a line given by 

2803: \beql{282}

2804: x\cos{\theta}  +y \sin{\theta}  = m,

2805: \eeq

2806: and its associated curvature-center coordinates are 

2807: \beql{283}

2808: (\bar{b}, b, w_1, w_2)= (2m, 0,\cos{\theta}, \sin{\theta}).

2809: \eeq

2810:  Here the

2811: orientation is given by the convention that the  

2812: normal $(\cos{\theta},\sin{\theta})$ points inward. 

2813: 

2814: 

2815: The group M\"{o}b(2) is generated by 

2816: 

2817:  (1) translations $\ft_{z_0}= z + z_0$;

2818: 

2819:  (2) dilations $\fd_\lambda(z) = \lambda z$ with $\lambda \in \cc$, 

2820: $\lambda \neq 0$;  

2821: 

2822:  (3) the conjugation $\fc (z) = \bar{z}$;

2823: 

2824:  (4)  the inversion in the unit circle

2825: $\fj(z) = \frac{1}{\bar{z}} = \frac{z}{|z|^2}.$

2826: 

2827: 

2828: Given $\fg \in \mbox{M\"{o}b}(2)$,

2829: we will let $\tilde{\fg}$ denote the 

2830: corresponding action on  the curvature-center coordinates of an oriented

2831: circle. 

2832: The action of translation by $z_0 = x_0 + iy_0$ is

2833: \beql{284}

2834: \tilde{\ft}_{z_0}(\bar{b}, b, w_1, w_2) = 

2835: (\bar{b}+2w_1x_0+2w_2y_0+b(x_0^2+y_0^2), b, w_1 + bx_0, w_2 + by_0).

2836: \eeq

2837: The action of a dilation with $\lambda=r e^{i\theta}$, $(r >0)$ is given by 

2838: \beql{285}

2839: \tilde{\fd}_\lambda (\bar{b}, b, w_1, w_2) = 

2840: (r\bar{b}, \frac{b}{r},~ w_1\cos{\theta}-w_2\sin{\theta}, w_1\sin{\theta}+w_2\cos{\theta}).

2841: \eeq

2842: The action  of complex conjugation is

2843: \beql{286}

2844: \tilde{\fc}(\bar{b}, b, w_1, w_2) = (\bar{b}, b, w_1, - w_2).

2845: \eeq

2846: The action of inversion is

2847: $$

2848: \tilde{\fj}( \bar{b}, b, w_1, w_2) = (b, \bar{b}, w_1, w_2).

2849: $$

2850: All of these actions apply

2851: to ``circles at infinity'' and extend  to linear maps on the $4 \times 4$ 

2852: matrices $\bW_{\sD}$. 

2853: 

2854: The translation operation is given by right multiplication by

2855: the matrix 

2856: \beql{Gt0}

2857: \bV_{\ft_{z_0}}^{-1} := \left[

2858: \begin{array}{crrc}

2859: 1 &  0  &  0  &  0 \\

2860: x_0^2+y_0^2  &  1   &  x_0   & y_0 \\

2861: 2x_0 &  0   &  1   & 0 \\

2862: 2y_0 & 0 & 0 & 1 

2863: \end{array}

2864: \right],

2865: \eeq

2866: and one verifies  \eqref{N9.2} holds by direct computation.

2867: 

2868: 

2869: For the  dilation $\fd_\lambda$, with $\lambda = re^{i \theta}$

2870: $(r > 0)$  the right action is by the matrix

2871: \beql{Gdlambda}

2872: \bV_{\fd_\lambda}^{-1} :=\left[

2873: \begin{array}{cccc}

2874: r         &   0            &     0          & 0 \\

2875: 0         & 1/r            &     0          & 0 \\

2876: 0         & 0              &  \cos{\theta}  & \sin{\theta}  \\

2877: 0         & 0              & -\sin{\theta}  & \cos{\theta}  

2878: \end{array} 

2879: \right].

2880: \eeq

2881: For complex conjugation

2882: $\fc$, the  right action is by the matrix 

2883: \begin{equation}

2884: \bV_{\fc}^{-1} = \bV_{\fc} 

2885: :=\left[

2886: \begin{array}{rrrr}

2887: 1         &  0             &     0          & 0 \\

2888: 0         &  1             &     0          & 0 \\

2889: 0         &  0              &    1          & 0  \\

2890: 0         &  0              &    0          & -1 

2891: \end{array} 

2892: \right].

2893: \end{equation}

2894: For  the inversion $\fj_C$ in the unit circle, the permutation matrix 

2895: \begin{equation}

2896: \bV_{\fj_C}^{-1} =\bV_{\fj_C}= P_{(12)} = \left[

2897: \begin{array}{cccc}

2898: 0         &   1            &     0          &    0 \\

2899: 1         &   0            &     0          &    0 \\

2900: 0         &   0            &     1          &    0  \\

2901: 0         &   0            &     0          &    1  

2902: \end{array} 

2903: \right].

2904: \end{equation}

2905: 

2906: It is easy to verify that the above matrices are all in $Aut(Q_W)^{\uparrow}$, 

2907: so that the map  so far defines  a homomorphism of $\mbox{M\"{o}b}(2)$

2908: into $Aut(Q_W)^{\uparrow}\simeq O(3, 1)^{\uparrow}$,

2909: identified with the  isochronous Lorentz group.

2910: The group $\mbox{M\"{o}b}(2)$ acts simply transitively on

2911: ordered Descartes configurations, as observed by Wilker

2912: \cite[Theorem 3, p. 394]{Wi81},

2913: and the group $Aut(Q_{W})$ acts simply transitively on

2914: ordered, oriented Descartes configurations by Theorem~\ref{th33}.

2915: Because  $Aut(Q_W)^{\uparrow}$ is of index $2$ in $Aut(Q_{W}) \simeq O(3,1)$,

2916: we conclude that the map so far defines  an isomorphism 

2917: of $\mbox{M\"{o}b}(2)$  onto $Aut(Q_W)^{\uparrow}$.

2918: 

2919: To complete the proof, we define  the action of $-\bI$ to be 

2920: \begin{equation}~\label{flip}

2921: (\bV_{-\bI})^{-1} = \bV_{-\bI} = - \bI.

2922: \end{equation}

2923: It has the effect of reversing  (total) orientation of the 

2924: Descartes configuration, and does not correspond to 

2925: a conformal transformation. Since $-\bI \notin Aut(Q_W)^{\uparrow}$,

2926: adding it gives the desired isomorphism of

2927: $GM(2)$ onto $Aut(Q_{W})$. $~~~\bsq$ \\

2928: 

2929: In   terms of  the natural coordinates

2930: on $\mbox{M\"{o}b}(2)_{+} \simeq PSL(2, \CC)$ the homomorphism

2931: $\pi(\cdot)$ given in Theorem~\ref{th91} 

2932: is a non-linear map. This can be clearly seen

2933: in \eqn{Gdlambda}, where both $\bU_{\fd_\lambda}$ and

2934: $\bU_{\fd_\lambda}^{-1}$ are nonlinear functions of the

2935: coordinates of

2936: \beql{N817a}

2937: {\fd}_{\lambda} = \left[ 

2938: \begin{array}{cc}

2939: \sqrt{\lambda} & 0 \\

2940: 0  & \frac{1}{\sqrt{\lambda}} 

2941: \end{array} \right] \in SL(2, \CC).

2942: \eeq

2943: Here the  two choices of  $\pm \sqrt{\lambda}$ give a well defined

2944: matrix modulo $\pm I$.

2945: 

2946: We now return to the data in  Table ~\ref{table1} 

2947: giving  the isomorphisms of $Aut(Q_{W})$ and its

2948: subgroups to the Lorentz group $O(3,1)$ and its

2949: two subgroups 

2950: $O(3,1)^{\uparrow}$ the

2951: orthochronous Lorentz group, and $O(3,1)_{+}^{\uparrow}$

2952: the proper orthochronous Lorentz group, 

2953: using  the terminology

2954: of Wilker ~\cite{Wi81}. An {\em orthochronous Lorentz transformation}

2955: is one that preserves the arrow of time; Table ~\ref{table1} shows

2956: this corresponds to preserving total orientation of a 

2957: Descartes configuration.

2958: The set of isomorphisms

2959: given by the three horizontal arrows

2960: on the right in Table~\ref{table1}

2961: are obtained by any fixed  choice of matrix $\bA$

2962: that intertwines $Q_{W}$ and $Q_{\sL}$ by

2963: $\bQ_{W} = \bA^T \bQ_{\sL} \bA$, in which case the 

2964: isomorphism is $Aut(Q_{W}) = \bA^{-1} O(3,1) \bA$

2965: sending $\bV \mapsto \bA \bV \bA^{-1}$. 

2966: To preserve the underlying rational structure

2967: of these forms, one must choose $\bA$ to be a

2968: rational matrix, for example the integral matrix

2969: given in \eqn{N333b}.

2970: 

2971: 

2972: There  is a different  choice of intertwining

2973: matrix, for the rightmost horizontal rows in

2974: Table A-1, which will allow us  to make the composed horizontal

2975: maps in Table \ref{table1} from the M\"{o}bius

2976: group to the Lorentz groups compatible with

2977: the framework used by Wilker \cite{Wi81}.

2978: In  \cite[Theorem 10]{Wi81} Wilker

2979: gave an 

2980: explicit isomorphism  of $PSL(2, \CC)$ onto 

2981: the proper orthochronous Lorentz group $O(3,1)_{+}^{\uparrow}$

2982: It maps

2983: \footnote{We have permuted the first and last row and column

2984: of Wilker's result because he uses the Lorentz form

2985: $w^2 + x^2+y^2-z^2$.}

2986: $$

2987: \pm \left[ \begin{array}{rr}

2988: a & b \\

2989: c & d 

2990: \end{array}

2991: \right] 

2992: \longmapsto

2993:  \left[

2994: \begin{array}{rrrr}

2995: \frac{1}{2}\left(|a|^2 + |b|^2 \right. & \Im(a\bar{c} + b \bar{d}) & 

2996: \frac{1}{2}\left (|a|^2 + |b|^2 \right. &  \Re(a\bar{c} + b\bar{d}) \\

2997: \left. +|c|^2 + |d|^2\right ) & &  \left. - |c|^2-|d|^2 \right)& \\

2998: \Im(-a\bar{b} - c\bar{d}) & \Re(a\bar{d} - b \bar{c}) &  \Im(-a\bar{b} +

2999: c\bar{d}) &  \Im(-a\bar{d} + b \bar{c}) \\

3000: \frac{1}{2}\left(|a|^2 - |b|^2 \right. &  \Im(a\bar{c}-b\bar{d}) & 

3001: \frac{1}{2}\left(|a|^2 -|b|^2 \right. &  \Re(a\bar{c}-b \bar{d}) \\

3002: \left. +|c|^2 - |d|^2\right)& & \left. - |c|^2 + |d|^2\right) & \\

3003: \Re(a\bar{b}+c\bar{d}) & \Im(a\bar{d}+ b \bar{c}) &  \Re(a\bar{b}-c\bar{d}) &

3004:  \Re(a\bar{d} + b \bar{c}) 

3005: \end{array}

3006: \right].

3007: $$

3008: One can identify this map  with the composed map in 

3009: the last row of Table A-1 

3010: from the first column  to the last column, if one 

3011: chooses  the map $Aut(Q_{W}) \to Aut(Q_{\sL})$

3012: to be  $\bV \mapsto \bZ \bV \bZ^{-1}$

3013: for  a particular intertwining matrix $\bZ$,

3014: which  is unique up to multiplication

3015: by $\pm \bI$. It has irrational entries, and is 

3016: \beql{convert}

3017: \bZ := \pm \sqrt{2} \left[

3018: \begin{array}{cccc}

3019: 1         &   1            &     0          &    0 \\

3020: 0         &   0            &     0          &    -1 \\

3021: 1         &   -1           &     0          &    0  \\

3022: 0         &   0            &     -1         &   0  

3023: \end{array} 

3024: \right].

3025: \eeq

3026: 

3027: 

3028: %(It is not the same as  \eqn{N333c}.)

3029: %

3030: % [[Warning: Wilker [p. 388] uses Lorentz form 

3031: %  U * V = U_1V_1 +...+U_{n+1}V_{n+1}- U_{n+2}V_{n+2}

3032: %  so a permutation exchanging first and last row and column was needed

3033: %  to convert his map to our Lorentz form.]]

3034: %

3035: %*****************************************************************************

3036: %

3037: % SECTION 10. Appendix B. Schottky Group Viewpoint

3038: %

3039: %*****************************************************************************

3040: %

3041: \newpage

3042: \section{Appendix B.  Groups in Hyperbolic $3$-Space associated 

3043: to Apollonian Packings} 

3044: \setcounter{equation}{0}

3045: 

3046: In this appendix we  describe two different discrete

3047: groups  of isometries of hyperbolic $3$-space $\HH^3$

3048: associated to an Apollonian packing. We specify these

3049: groups as groups of M\"{o}bius transformations of $\hat{\cc}$.

3050: 

3051: Any M\"{o}bius transformation of $\hat{\cc} = \hat{\rr}^2$

3052: has a unique Poincare lift to an isometry of hyperbolic $3$-space

3053: $\HH^3$, viewed as the upper half-space of 

3054: $\hat{\rr}^3 = \rr^3 \cup \{\infty\}$,

3055: as described in Beardon~\cite[Sect. 3.3]{Be83}. Here 

3056: $\hat{\cc}$ is identified with the ideal boundary of $\hat{\rr}^3$.

3057: In this way any  group of M\"{o}bius transformations acting on

3058: $\hat{\cc}$  lifts to a group of hyperbolic isometries.

3059: 

3060: A {\em classical Schottky group} is a group 

3061: $\Gamma = \langle \fg_1, ...\fg_n \rangle$ of  

3062: M\"{o}bius transformations 

3063: associated to $2n$ circles in the plane $\hat{\cc}$ having

3064: disjoint interiors, but possibly touching on their boundaries.

3065: The group has $n$ generators $\fg_j$,   in which the 

3066: $j$-th generator of the group maps the exterior of circle $C_{2j-1}$

3067: onto the interior of circle $C_{2j}$.

3068: Schottky groups consist entirely of

3069: holomorphic M\"{o}bius transformations.

3070: (Note: knowing a single $\fg_j$ alone does not determine the

3071: circles $C_{2j-1}$ and $C_{2j}$ uniquely.)

3072: 

3073: Mumford, Series and Wright \cite[Chapter 7]{MSW02}

3074: observe that given an  Apollonian circle packing $\sP_{\sD}$

3075: generated by a Descartes configuration $\sD$,

3076: there is a Schottky group $\Gamma (\sD)$ in 

3077: hyperbolic $3$-space $\HH^3$ whose limit

3078: set is exactly the limit set $\Lambda_{\sD}$ 

3079: of the Apollonian packing $\sP_{\sD}$.

3080: The Schottky group $\Gamma(\sD) $ has two generators

3081: $\bP_1, \bP_2$, involving maps between

3082: two pairs of circles in the dual Descartes

3083: configuration

3084: $\sD^{\perp}$, specified by

3085: the ordering as $A, a, B, b$,

3086: say. It takes the exterior of circle $A$ to the

3087: interior of circle $a$ and the exterior of circle $B$ to

3088: the interior of circle $b$.  

3089: 

3090: We treat the case they picture \cite[Figure 7.3]{MSW02}.

3091:  where the Schottky limit

3092: set is $\Lambda(\sD_0)$,

3093: for  the (ordered) Descartes

3094: configuration $\sD_0$ having coordinates

3095: $$

3096: \bW_{\sD_0}= \left[

3097: \begin{array}{rrrr}

3098: 1  & -1 & 0 & 0 \\

3099: 0  & 2 & 1 & 0 \\

3100: 0  & 2 & -1 & 0 \\

3101: 1  & 3 &  0 & -2

3102: \end{array}\right].

3103: $$

3104: This configuration is  pictured in Figure \ref{B1-special}, with part (a)

3105: indicating curvatures and part (b) shows the ordering of  circles,

3106: used later in labeling  M\"{o}bius  group generators $\fs_i$.

3107: 

3108: %

3109: % fig B-1 here 

3110: %

3111: %

3112: \begin{figure}[htbp]

3113: \centerline{\epsfxsize=4.0in \epsfbox{B1-special.ps}}

3114: \caption{Special Descartes configuration $\sD_0$}~\label{B1-special}

3115: \end{figure}

3116: 

3117: The Schottky group operations are associated to the dual

3118: Descartes configuration $\sD_0^{\perp}$, which  has 

3119: augmented curvature-center coordinates

3120: $$

3121: \bW_{{\sD}_0^{\perp}}= \left[

3122: \begin{array}{rrrr}

3123: 0 & 0 & 0 & 1 \\

3124: 1 & 1 & -1 & -1 \\

3125: 1 & 1 & 1 & 1 \\

3126: 0 & 4 & 0 & -1 \\

3127: \end{array}\right].

3128: $$

3129: It is pictured in Figure \ref{B2-dual}, with (a) indicating

3130: curvatures and (b) the labeling of circles used in 

3131: the Schottky group.

3132: %

3133: % fig B-2 here 

3134: %

3135: %

3136: \begin{figure}[htbp]

3137: \centerline{\epsfxsize=4.0in \epsfbox{B2-dual.ps}}

3138: \caption{Dual Descartes configuration $\sD_0^{\perp}$}~\label{B2-dual}

3139: \end{figure}

3140: 

3141: 

3142: 

3143: In Figure \ref{B2-dual}  the Schottky group generator 

3144: $\bP_1$ maps the exterior of circle $A$ to the

3145: interior of circle $a$, and $\bP_2$ maps the exterior

3146: of circle $B$ to the interior of circle $b$.

3147: We let $\Gamma_{S}(\sD)$ denote the Schottky group

3148: they generate.

3149: The limit set  $\Lambda_{\Gamma_{S}(\sD)}$ of

3150: this Schottky group is the

3151: closed set in  $\PP^1(\CC)$ obtained by removing 

3152: the interiors of all circles in the Apollonian packing

3153: associated to $\sD$. This is exactly

3154: the residual set of the Apollonian packing,

3155:  see Theorem~\ref{th41b} for its properties. It is

3156: a fractal set of Hausdorff dimension about $1.3$.

3157: 

3158: The precise result is as follows.

3159: 

3160: 

3161: \begin{prop}~\label{thB1}

3162: The Schottky group $\Gamma = \langle \bP_1, \bP_2 \rangle$ given 

3163: by the parabolic generators

3164: \beql{1001}

3165: \bP_1 = \left[ \begin{array}{cc}

3166: 1 -i  & 1  \\

3167: 1  & 1 +i

3168: \end{array}\right]

3169: ~~~~\mbox{and}~~~~

3170: \bP_2 = \left[ \begin{array}{cc}

3171: 1  & 0  \\

3172: -2i  & 1

3173: \end{array}\right]

3174: \eeq

3175: has a limit set $\Lambda_{\Gamma}$ which is the complement

3176: of the interiors of the circles in the Apollonian packing $\sP_{\sD_0}$

3177: associated to the Descartes configuration $\sD_0$ above. 

3178: \end{prop}

3179: 

3180: This limit set,  ``the glowing gasket'',  is illustrated in Figure 7.3 in 

3181: Mumford, Series and Wright \cite[Chapter 7]{MSW02}. 

3182: 

3183: The Schottky group $\Gamma_{S}(\sD_0)$ is very special;

3184: it admits no deformations other than conjugacy, and

3185: under conjugacy $\fg \Gamma_{S}(\sD_0)\fg^{-1} =\Gamma_{S}(\sD) $

3186: with $\sD = \fg(\sD_0)$.

3187:  Recall that

3188: a  matrix $\bM \in PSL(2, \CC)$ is

3189: {\em parabolic} if $Trace(\bM) = \pm 2$; its 

3190: action on the Riemann sphere has a single fixed point.

3191: Here $\bP_1, \bP_2$ are parabolic, as is

3192: \beql{1001b}

3193: [\bP_1, \bP_2] :=  \bP_1 \bP_2 \bP_1^{-1} \bP_2^{-1} =  

3194: \left[ \begin{array}{cc}

3195: -1 -2i  & 2i  \\

3196: -2i  & -1 +2i

3197: \end{array}\right].

3198: \eeq

3199: The Schottky group $\Gamma_{S}(\sD_0)$ is

3200: characterized up to conjugacy

3201: by the property that it is generated by 

3202: two elements $\bP_1, \bP_2$

3203: which are parabolic,  such that their commutator 

3204: $[\bP_1, \bP_2] = \bP_1 \bP_2 \bP_1^{-1} \bP_2^{-1}$ is

3205: also parabolic (\cite[pp. 207--208]{MSW02}).

3206: 

3207: Now consider a the  group 

3208: $G_{\sA}({\sD_0}) = \langle \fs_1, \fs_2, \fs_3, \fs_4 \rangle$

3209: of M\"{o}bius transformations 

3210: corresponding to the Apollonian group for

3211: the Descartes configuration $\sD_0$, see \eqn{4311}.

3212: Here $\bs_i$ is inversion in the i-th circle of the

3213: dual Descartes configuration $\sD^{\perp}$, which is the 

3214: circle passing through the three intersection points in $\sD$

3215: not touching $C_i$.

3216: The group $G_{\sA}({\sD_0})$ lifts to a group $\Gamma_{A}(\sD_0)$

3217: of isometries of $\HH^3$. 

3218: We describe $G_{\sA}({\sD_0})$  for  the case that

3219: $\sD_0$ is the Descartes configuration in Figure~\ref{B1-special}

3220: with circles numbered as in Figure~\ref{B1-special}(b) ; 

3221: all other

3222: $G_{\sA}(\sD)$ are related to it by conjugacy in

3223: the M\"{o}bius group.

3224: An inversion in a circle $C$ of radius $r$

3225: centered at $\bz_0$ is the anti-holomorphic map 

3226: $$

3227: \fj_{C}(z)= z_0 + \frac{r^{2}}{\overline{z} - \overline{z_0}}

3228: = \frac{ z_0 \bar{z} + r^{2} - |z_0|^2}{\bar{z} - \bar{z}_0}.

3229: $$

3230: We can write  the 

3231: group generators  as $\fs_j =\fp_j \circ \fc$, 

3232: where $\fp_j \in PSL(2, \cc)$  is holomorphic and $\fc$ denotes

3233: complex conjugation, with

3234: $$

3235: \fp_1 = \left[ \begin{array}{cc}

3236: 1-i & -i \\

3237: i & 1+i

3238: \end{array}\right],~~~

3239: \fp_2 = \left[ \begin{array}{cc}

3240: 1+i  & -i \\

3241: i &  1-i

3242: \end{array}\right],~~~

3243: \fp_3 = \left[ \begin{array}{cc}

3244: 1 & 0 \\

3245: 4i  & 1

3246: \end{array}\right],~~~

3247: \fp_4 = \left[ \begin{array}{cc}

3248: 1 & 0 \\

3249: 0 & 1

3250: \end{array}\right].

3251: $$

3252: Since $\fp_4 = \bI$,

3253: the  subgroup of holomorphic elements of $G_{\sA}(\sD)$ is

3254: \begin{eqnarray}

3255: G_{\sA}^2(\sD) &= & \langle \fp_1\overline{\fp}_2, \fp_1\overline{\fp}_3, 

3256: \fp_1 \overline{\fp}_4,

3257: \fp_2 \overline{\fp_3}, \fp_2\overline{\fp_4}, \fp_3\overline{\fp_4},

3258:  \fp_2\overline{\fp}_1, \fp_3\overline{\fp}_1, \fp_4 \overline{\fp}_1,

3259: \fp_3 \overline{\fp_2}, \fp_4\overline {\fp}_2, \fp_4\overline{\fp_3}

3260:  \rangle \nonumber \\

3261: &=& \langle \fp_1, \fp_2, \fp_3 \rangle,

3262: \end{eqnarray}

3263: where we used the fact that $\overline{\fp_j} = \fp_j^{-1}$.

3264: 

3265: There are clearly relations

3266: between the 

3267: Schottky  group $\Gamma_{S}(\sD_0)$ and the 

3268:  group $G_{\sA}^2(\sD_0)$ of holomorphic elements of $G_{\sA}(\sD)$

3269: Indeed the group $G_{\sA}^2(\sD_0)$ is generated by three parabolic

3270: elements $\fp_1, \fp_2, \fp_3$, and these are related to the three

3271: parabolic elements $P_1, P_2, [P_1, P_2]$ 

3272: in $\Gamma_{S}(\sD)$ by the relations

3273: $$

3274: \fp_3 = (P_2)^{-2},~~~~~\fp_1^{-1}\fp_2= -(P_1)^{-2},~~~~~

3275: \fp_2^2 = -[P_1, P_2].

3276: $$

3277: However these groups are not exactly the same. It remains to determine

3278: a more precise relation between them,  explaining how they are 

3279: give rise to the  same Apollonian packing. Note that both these

3280: groups lifted to discrete isometry groups of $\HH^3$

3281: have fundamental domains in $\HH^3$ of infinite volume. 

3282: 

3283: \clearpage

3284: \begin{thebibliography}{99}

3285: 

3286: \bibitem{AS97}

3287: D. Aharonov and K. Stephenson, 

3288: Geometric sequences of discs in the

3289: Apollonian packing. 

3290: Algebra i Analiz {\bf 9} (1997), No. 3, 104--140.

3291: [English version: St. Petersburg Math. J. {\bf 9} (1998), 509--545.]

3292: 

3293: \bibitem{Ba04}

3294: P. Bahls,

3295: Relative hyperbolicity and right-angled Coxeter groups,

3296: eprint: {\tt arXiv:math.GR/041280 21 Jan 2004}.

3297: 

3298: \bibitem{Be83}

3299: A. F. Beardon,

3300: {\em The Geometry of Discrete Groups,}

3301: Springer-Verlag: New York 1983.

3302: 

3303: \bibitem{Ber87}

3304: M. Berger,

3305: {\em Geometry II},

3306: Springer-Verlag: Berlin 1987.

3307: 

3308: \bibitem{Bo73}

3309: D. W. Boyd,

3310: The osculatory packing of a three-dimensional sphere.

3311: Canadian J. Math. {\bf 25} (1973), 303--322.

3312: 

3313: \bibitem{Boyd73}

3314: D. W. Boyd, 

3315: The residual set dimension of the Apollonian packing.

3316: Mathematika {\bf20} (1973), 170--174.

3317: % showed exponent=hausdorff dim for Apollonian packing.

3318: 

3319: \bibitem{Bo73b}

3320: D. W. Boyd,

3321: Improved bounds for the disk-packing constant,

3322: Aequations Math. {\bf 9} (1973), 99--106.

3323: 

3324: \bibitem{Bo74}

3325: D. W. Boyd,

3326: A new class of infinite sphere packings,

3327: Pacific J. Math. {\bf 50} (1974), 383--398.

3328: 

3329: \bibitem{Bo82}

3330: D. W. Boyd,

3331: The sequence of radii of the Apollonian packing.

3332: Math. Comp. {\bf 39} (1982), 249--254.

3333: 

3334: \bibitem{Br85}

3335: R. W. Brooks,

3336: The spectral geometry of the Apollonian packing,

3337: Comm. Pure. Appl. Math. {\bf 38} (1985), 359--366.

3338: 

3339: \bibitem{Cas78}

3340: J. W. S. Cassels,

3341: {\em Rational Quadratic Forms},

3342: Academic Press: New York 1978.

3343: 

3344: \bibitem{Con97}

3345: J. H. Conway, with F. Fung,

3346: {\em The sensual (quadratic) form,}

3347: Carus Monograph No. 26, Math. Assoc. America,

3348: Washington DC, 1997.

3349: 

3350: \bibitem{Cox68}

3351: H. S. M. Coxeter, 

3352: The problem of Apollonius. 

3353: Amer. Math. Monthly {\bf 75} (1968), 5--15. 

3354: 

3355: \bibitem{Cox69}

3356: H. S. M. Coxeter,

3357: {\em Introduction to Geometry, Second Edition},

3358: John Wiley and Sons, New York, 1969.

3359: 

3360: \bibitem{Cox68b}

3361: H. S. M. Coxeter,

3362: Loxodromic sequences of tangent spheres. 

3363: Aequationes Mathematicae {\bf 1} (1968), 104--121.

3364: 

3365: \bibitem{Cox97}

3366: H. S. M. Coxeter,

3367: Numerical distances among the spheres in a loxodromic sequence. 

3368: The Mathematical Intelligencer {\bf 19} (1997), 41--47.

3369: 

3370: \bibitem{Des01}

3371: R. Descartes.

3372: {\em Oeuvres de Descartes, Correspondence IV},

3373: (C. Adam and P. Tannery, Eds.), Paris: Leopold Cerf 1901.

3374: 

3375: \bibitem{Fa85}

3376: K. J. Falconer,

3377: {\em The Geometry of Fractal Sets}. 

3378: Cambridge Tracts in Math., vol. 85, Camb. Univ. Press, Cambridge, 

3379: 1985.

3380: 

3381: \bibitem{Go37}

3382: T. Gossett,

3383: The Kiss Precise,

3384: Nature {\bf 139}(1937), 62.

3385: 

3386: \bibitem{Go37b}

3387: T. Gossett,

3388: The Hexlet,

3389: Nature {\bf 139} (1937), 251.

3390: 

3391: 

3392: \bibitem{GLMWY22}

3393: R. L. Graham, J. C. Lagarias, C. L. Mallows, A. Wilks and C. Yan,

3394: Apollonian circle packings: geometry and group theory

3395: II. Super-Apollonian group and integral packings,

3396: eprint: {\tt arXiv math.MG/0010302}

3397: 

3398: \bibitem{GLMWY23}

3399: R. L. Graham, J. C. Lagarias, C. L. Mallows, A. Wilks and C. Yan,

3400: Apollonian circle packings: geometry and group theory

3401: III. Higher dimensions, eprint:{\tt arXiv math.MG/0010324}.

3402: 

3403: \bibitem{GLMWY2}

3404: R. L. Graham, J. C. Lagarias, C. L. Mallows, A. Wilks and C. Yan,

3405: Apollonian circle packings: number theory, J. Number Theory

3406: {\bf 100} (2003), 1--45.

3407: [eprint: {\tt arXiv math.NT/009113}]

3408: 

3409: \bibitem{He21}

3410: T. E. Heath,

3411: {\em A History of Greek Mathematics, Volume II. From Aristarchus

3412: to Diophantus}, 

3413: Dover, New York 1981. (Original: Clarendon Press, Oxford 1921).

3414: 

3415: 

3416: \bibitem{Hi67}

3417: K.E. Hirst, 

3418: The Apollonian packing of circles. 

3419: J. Lond. Math. Soc., {\bf 42}

3420: (1967), 281--291. 

3421: 

3422: 

3423: \bibitem{KS43}

3424: E. Kasner and F. Supnick,

3425: The Apollonian packing of circles.

3426: Proc. Nat. Acad. Sci. USA {\bf 29} (1943), 378--384.

3427: 

3428: \bibitem{Lac86}

3429: R. Lachlan,

3430: On systems of circles and spheres,

3431: Phil. Trans. Roy. Soc. London, Ser. A 

3432: {\bf 177} (1886), 481--625.

3433: 

3434: \bibitem{LMW02}

3435: J. C. Lagarias, C. L. Mallows and A. Wilks,

3436: Beyond the  Descartes circle theorem, 

3437: Amer. Math. Monthly {\bf 109} (2002), 338--361.

3438: [eprint: {\tt arXiv math.MG/0101066}]

3439: 

3440: \bibitem{Lar66}

3441: D. G. Larman,

3442: On the exponent of convergence of a packing of spheres,

3443: Mathematika {\bf 13} (1966), 57--59.

3444: 

3445: \bibitem{M82}

3446: B. B. Mandelbrot,

3447: {\em The Fractal Geometry of Nature},

3448: Freeman: New York, 1982.

3449: 

3450: \bibitem{Max82}

3451: G. Maxwell,

3452: Sphere packings and hyperbolic reflection groups.

3453: J. Algebra {\bf 79} (1982), 78--97.

3454: 

3455: \bibitem{Me66}

3456: Z. A. Melzak,

3457: Infinite packings of disks.

3458: Canad. J. Math. {\bf 18} (1966), 838--853.

3459: 

3460: \bibitem{MSW02}

3461: D. Mumford, C. Series, and D. Wright,

3462: {\em Indra's Pearls: The Vision of Felix Klein},

3463: Cambridge U. Press: Cambridge 2002.

3464: 

3465: \bibitem{No03}

3466: S. Northshield,

3467: On Apollonian circle packings,

3468: preprint.

3469: 

3470: \bibitem{Pe67}

3471: D. Pedoe,

3472: On a theorem in geometry,

3473: Amer. Math. Monthly {\bf 74} (1967), 627--640.

3474: 

3475: \bibitem{RS87}

3476: B. Rodin and D. Sullivan,

3477: The convergence of circle packings to the Riemann mapping. 

3478: J. Differential Geometry {\bf 26} (1987), 349--360. 

3479: 

3480: 

3481: %\bibitem{Ro98}

3482: %T. Rothman, 

3483: %Japanese temple geometry. 

3484: %Scientific American, May 1998, 84--91.

3485: 

3486: %\bibitem{Ru70}

3487: %W. R\"uhl,

3488: %{\em The Lorentz group and harmonic analysis},

3489: %W. A. Benjamin: New York 1970.

3490: 

3491: %\bibitem{S98}

3492: %N. J. A. Sloane,

3493: %The on-line encyclopedia of integer sequences. \newline

3494: %(URL is http://www.research.att.com/\~{}njas/sequences/index.html)

3495: 

3496: \bibitem{Sod36}

3497: F. Soddy, 

3498: The Kiss Precise. 

3499: Nature {\bf 137} (1936), 1021. 

3500: 

3501: \bibitem{Sod37}

3502: F. Soddy,

3503: The bowl of integers and the Hexlet,

3504: Nature {\bf 139} ( 1937), 77--79.

3505: 

3506: 

3507: \bibitem{So92}

3508: B. S\"oderberg,

3509: Apollonian tiling, the Lorentz group, and regular trees.

3510: Phys. Rev. A {\bf 46} (1992), No. 4, 1859--1866.

3511: 

3512: \bibitem{TD94}

3513: P. B. Thomas and D. Dhar,

3514: The Hausdorff dimension of the Apollonian packing of circles.

3515: J. Phys. A: Math. Gen. {\bf 27} (1994), 2257--2268.

3516: 

3517: \bibitem{Tri84}

3518: C. Tricot, 

3519: A new proof for the residual set dimension of the Apollonian packing, 

3520: Math. Proc. Cambridge Phil. Soc. {\bf 96} (1984), 413--423.        

3521: 

3522: %\bibitem{Wi67}

3523: %J. B. Wilker,

3524: %Open disk packings of a disk. 

3525: %Canad. Math. Bull., {\bf 10} (1967), 395--415.

3526: 

3527: \bibitem{Wi81}

3528: J. B. Wilker,

3529: Inversive Geometry,

3530: in: {\em The Geometric Vein}, (C. Davis, B. Gr\"unbaum, F. A. Sherk, Eds.),

3531: Springer-Verlag: New York 1981, pp. 379--442.

3532: 

3533: \end{thebibliography}

3534: 

3535: {\tt

3536: \begin{tabular}{lllll}

3537: email: & graham@ucsd.edu \\

3538: & lagarias@umich.edu \\

3539:   & colinm@research.avayalabs.com \\

3540:  & allan@research.att.com \\

3541: & cyan@math.tamu.edu

3542: \end{tabular}

3543:  }

3544: 

3545: 

3546:        

3547: \end{document}

3548: 

3549: 

3550: 

3551: 

3552: 

3553: 

3554: 

3555: 

3556: 

3557: 

3558: 

3559: 

3560: 

3561: