1: % revision as suggested by referee of DCG, March 2, 2005. 

2: %

3: % Part II, 9 jun 04 (rlg corrections)

4: % 8 jun 04 

5: % 5 jun 04 ( cyan Thm 5.2 replace)

6: % new file from scratch 2 may 04 [April 15 last version]

7: %

8: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

9: % 11 Figure files 

10: %

11: %  0.0.1.1.ps 1.2.2.3.ps 6.11.14.15.ps

12: %  super.ps   

13: %  all6.ps all8.ps all9.ps 

14: %  0mod1.ps

15: %  1mo2.ps 2mod4.ps 0.mod4.ps

16: %   fig1.1a, 1-std.ps, 6-std.p

17: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%  New results added since 2001 version:

18: %

19: %    Theorem 3.1, Theorem 3.2 (geometric ``packing''

20: %    Theorem 4.1 (restated, not new otherwise)

21: %    Theorem 5.1 (restated, not new otherwise)

22: %    Theoem 6.2, Theorem 6.3

23: %    Theorem 8.1 (CLM)

24: %    

25: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

26: \documentclass[11pt]{article}

27: %\usepackage{showlabels}

28: 

29: \usepackage{epsfig,amsmath,latexsym}

30: \usepackage{amsfonts,amssymb}

31: %\usepackage{diagrams}

32: 

33: 

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

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

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

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

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

39: 

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

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

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

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

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

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

46: 

47: 

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

49: 

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

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

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

53: 

54: 

55: \newcommand{\ep}{\epsilon}

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

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

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

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

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

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

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

63: 

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

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

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

67: \newcommand{\lf}{\lfloor}

68: \newcommand{\rf}{\rfloor}

69: \newcommand{\lc}{\lceil}

70: \newcommand{\rc}{\rceil}

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

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

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

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

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

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

77: 

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

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

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

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

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

83: 

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

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

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

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

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

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

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

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

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

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

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

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

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

97: 

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

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

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

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

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

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

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

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

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

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

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

109: 

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

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

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

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

114: \newcommand{\bG}{{\mathbf G}}

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

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

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

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

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

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

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

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

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

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

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

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

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

128: \newcommand{\asc}{{\bf asc}}

129: 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

144: 

145: 

146: \makeatletter

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

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

149:      \def\@svsec{}\else

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

151:      \@tempskipa #5\relax

152:       \ifdim \@tempskipa>\z@

153:         \begingroup #6\relax

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

155:         \endgroup

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

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

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

159:                     #7}\else

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

161:                       {#7}\addcontentsline

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

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

164:                        #7}}\fi

165:      \@xsect{#5}}

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

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

168: 

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

170:      \def\batchfile{here.doc}

171:      \input docstrip

172:      \preamble

173: 

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

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

176: 

177:      \endpreamble

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

179:      \endinput

180: \fi

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

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

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

184: \let\@Hxfloat\@xfloat

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

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

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

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

189: \ignorespaces}

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

191: 

192: \makeatother

193: 

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

195: 

196: \catcode`\@=11

197: %\renewcommand{\section}{

198: %        \setcounter{equation}{0}

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

200: %        -.2ex}{2.3ex plus .2ex}{\large\bf}

201: %        }

202: 

203: \catcode`@=12

204: 

205: 

206: \thispagestyle{empty}

207: 

208: \begin{document}

209: 

210: 

211: 

212: % TITLE PAGE and ABSTRACT

213: 

214: \begin{center}

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

216: {\Large {\bf II. Super-Apollonian Group and Integral Packings}}\\

217: \vspace{1.5\baselineskip}

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

219: Department  of Computer Science and Engineering\\

220: University of California at San Diego, 

221: La Jolla, CA 92093-0114 \\

222: \vspace*{1.5\baselineskip}

223: 

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

225: Department of Mathematics \\

226: University of Michigan, 

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

228: \vspace*{1.5\baselineskip}

229: 

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

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

232: \vspace*{1.5\baselineskip}

233: 

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

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

236: \vspace*{1.5\baselineskip}

237: 

238: {\em Catherine H. Yan}

239: \footnote{

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

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

242: Technology, China.}\\

243: Department of Mathematics \\

244: Texas A\&M University,

245: College Station, TX 77843-3368\\

246: \vspace*{1.5\baselineskip}

247: %\today

248: (March 10, 2005 version) \\

249: \vspace*{1.5\baselineskip}

250: {\bf ABSTRACT}

251: \end{center}

252: Apollonian circle packings arise by repeatedly filling the interstices

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

254: Such packings can be described in terms of the Descartes configurations

255: they contain, where a  

256: Descartes configuration is a set of four mutually tangent

257: circles in the Riemann sphere, having disjoint interiors.

258: Part I showed there exists a discrete group, the Apollonian group,

259: acting on a parameter space of 

260: (ordered, oriented) Descartes configurations, 

261: such that the Descartes configurations in a packing formed an

262: orbit under the action of this group. It is observed there

263: exist infinitely many types of integral

264: Apollonian  packings in which all circles had integer curvatures,

265: with the integral structure being related to the integral

266: nature of the Apollonian group. 

267: Here we consider the

268: action of a larger discrete group, the super-Apollonian

269: group, also having an integral structure,

270: whose orbits describe the Descartes quadruples of a geometric

271: object we call a 

272: super-packing. The circles in a super-packing never cross each other

273: but are nested to an arbitrary depth.

274: Certain Apollonian packings and super-packings

275: are strongly integral in the sense that the curvatures of all

276: circles are integral and the curvature$\times$centers of all

277: circles are integral. We show that (up to scale)

278: there are exactly $8$ different

279: (geometric) strongly integral super-packings, and that each

280: contains a copy of every integral Apollonian circle packing

281: (also up to scale). We show  that the super-Apollonian group

282: has finite volume in the group of all automorphisms of the

283: parameter  space of 

284: Descartes configurations, which is isomorphic

285: to the Lorentz group $O(3, 1)$.

286: 

287: \vspace*{1.5\baselineskip}

288: \noindent

289: Keywords: Circle packings, Apollonian circles, Diophantine equations,

290: Lorentz group, Coxeter group

291: 

292: \setcounter{page}{1}

293: 

294: 

295: %*****************************************************************************

296: %

297: % SECTION 1. 

298: %

299: %*****************************************************************************

300: %

301: \setlength{\baselineskip}{1.0\baselineskip}

302: \section{Introduction}

303: \setcounter{equation}{0}

304: 

305: 

306: Apollonian circle packings are arrangements of

307: tangent circles that arise by repeatedly filling the interstices

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

309: A set of four mutually tangent circles is called a

310: Descartes configuration. 

311: Part I studied Apollonian circle packings in terms of

312: the set of Descartes configurations that they contain.

313: It is observed that there exist  Apollonian circle packings 

314: that have a very strong integral structure, with all

315: circles in the packing having integer curvatures,

316: and rational centers, such that curvature$\times$center

317: is an integer vector. We termed these {\em strongly

318: integral} Apollonian circle packings. An example is 

319: the $(0,0,1,1)$ packing pictured in Figure~\ref{fig1},

320: with the two circles of radius $1$ touching at the origin, and with

321: two straight lines parallel to the $x$-axis.

322: 

323: %

324: % Figure 1  (0,0,1,1)

325: %

326: 

327: \begin{figure}[htbp]

328: \centerline{\epsfxsize=5.0in \epsfbox{0.0.1.1.ps}}~\label{fig1}

329: \caption{The integer  Apollonian packing $(0,0, 1,1)$}

330: \end{figure}

331: 

332: 

333: Part I gave an explanation for the existence of such

334: integral structures. This uses  a coordinate

335: system for describing all (ordered, oriented)

336: Descartes configurations $\sD$ in terms of their 

337: curvatures and centers, which forms a $4 \times 3$

338: {\em curvature-center coordinate}

339: matrix $\bM_{\sD}$, and a more detailed coordinate system,

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

341: using $4\times 4$ matrices $\bW_{\sD}$. 

342: The strongly integral property of a single Descartes configuration

343: is encoded in the integrality of the matrix $\bM_{\sD}$.

344: The set of all (geometric) Descartes configurations in an 

345: Apollonian packing can be described as a single  orbit of

346: a certain discrete group $\sA$ of $4 \times 4$ 

347: integer matrices; algebraically there are $48$  orbits

348: of ordered, oriented Descartes configurations giving rise to 

349: the same geometric packing, which  correspond to the 

350: 48 possible ways of ordering the circles and totally orienting the

351: configuration.

352: The integrality of the members

353: of $\sA$ is the source of the strong integrality

354: of some Apollonian circle packings. 

355: As a consequence of this group action, if  a single

356: Descartes configuration in the packing is strongly

357: integral, then they all are; hence  

358: every individual circle in the

359: packing is strongly integral. 

360: 

361: There are infinitely many distinct integral Apollonian

362: circle packings. Two more  of them are pictured in

363: Figures 2 and 3,  the $(-1, 2,2,3)$-packing

364: and the $(-6, 11, 14, 15)$-packing, respectively.

365:  Any integral packing

366: can be moved by a Euclidean motion so as to be

367: strongly integral, as will follow from results in

368: this paper.

369: 

370: 

371: %

372: % Figure 2  (-1,2,2,3)

373: %

374: 

375: \begin{figure}[htbp]

376: \centerline{\epsfxsize=3.0in \epsfbox{1.2.2.3.ps}}

377: \caption{The integer  Apollonian packing $(-1,2,2,3)$}~\label{fig2}

378: \end{figure}

379: %

380: % Figure 3 (-6, 11, 14, 15))

381: %

382: %

383: \begin{figure}[htbp]

384: \centerline{\epsfxsize=3.0in \epsfbox{6.11.14.15.ps}}

385: \caption{The integer  Apollonian packing $(-6,10,11,14)$}~\label{fig3}

386: \end{figure}

387: 

388: 

389: Part I introduced some further integral actions on

390: Descartes configurations, involving a ``duality''

391: operator $\bD$ leading to a dual Apollonian group $\sA^{\perp}$.

392: Combining this group with the Apollonian group

393: leads to a large group of integer $4 \times 4$ matrices,

394: the {\em super-Apollonian group} $\sA^{S}$, which 

395: acts on the set of all (ordered, oriented) Descartes

396: configurations. Part I showed that the super-Apollonian

397: group is a hyperbolic Coxeter group. 

398: It defined an {\em Apollonian super-packing}

399: to be an orbit  $\sA^{S}[\sD]$ of a single

400: Descartes configuration $\sD$ under this group. 

401: Such a super-packing is called  {\em integral}

402: if the initial Descartes configuration has integer

403: curvatures, and is  

404: {\em strongly integral} if the curvature-center coordinates

405:  $\bM_{\sD}$ of the initial Descartes configuration

406: are integral. These properties hold for all Descartes

407: configurations in the packing if they hold for one.

408: 

409:  In this paper we study the geometric 

410: structure  of super-packings,

411: and their integrality properties.

412: A  striking geometric fact is

413:  that the circles in a super-packing do not

414: overlap, as shown in \S3. Figure~\ref{fig4}  shows

415: circles of curvature at most $65$ in the 

416: super-packing generated from the $(0,0,1,1)$ 

417: configuration in Figure~\ref{fig1} above.

418: (The generating Descartes configuration is indicated

419: with slightly darker lines.)

420: Here the horizontal and vertical scales of the

421: figure are roughly from $-2.2 \le x \le 2.2$

422: and $-2.2 \le y \le 2.2$. This picture is 

423: representative of  the

424: whole super-packing; we show in \S6 that this

425: super-packing is periodic under the 

426: two-dimensional lattice generated

427: by the shifts

428: $x \to x+2$ and $y \to y+2$.  

429: 

430: %

431: % Figure 4  (superpacking to depth 20)

432: %

433: %

434: \begin{figure}[htbp]

435: \centerline{\epsfxsize=6.0in \epsfbox{super.ps}}

436: \caption{An initial part of the  $(0,0,1,1)$ super-packing

437: (square of sidelength $4.4$)}~\label{fig4}

438: \end{figure}

439: 

440: We cannot picture the whole super-packing because

441: the circles in it are dense in the plane; they nest inside

442: each other to an arbitrary depth. One can show that

443: the set of points lying in the interior of an infinite

444: nested sequence of such circles has full Lebesgue

445: measure. However there is interesting structure

446: visible if we turn up the scale of the circles.

447: In \S6 in Figure~\ref{fig8} we show all circles of curvature at

448: most 200 inside the super-packing above, inside

449: the unit square $0 \le x \le 1,~ 0 \le y \le 1$.

450: 

451: 

452: This paper is concerned with integrality 

453: properties of some super-packings, and we study successively

454: stronger integrality properties. In \S4 we require the

455: curvatures to be integral, in \S5 and \S6 both

456: the curvatures and curvature$\times$center must be 

457: integral, and in \S8 the augmented curvature-center

458: coordinates of all Descartes configurations in the 

459: super-packing must be integral. At this last level we

460: show that there are exactly $14$ such super-packings, viewed

461: as rigid geometric objects.

462: In \S2 we give a more detailed description of our results.

463: 

464: The general framework of this paper 

465: was developed by  the second author (JCL), who also did much of

466: the writing. This paper is an extensive

467: revision  of a preprint written in 2000, and adds new

468: results in \S3 and  \S6.

469: 

470: 

471: \paragraph {Acknowledgments.} The authors are grateful for

472: helpful comments from Andrew Odlyzko, Eric Rains,

473: Jim Reeds and  Neil Sloane during this work. They

474: also thank the  reviewer for incisive comments. \\

475: 

476: %*****************************************************************************

477: %

478: % SECTION 2. Summary

479: %

480: %*****************************************************************************

481: %

482: \section{Summary of Main Results}

483: \setcounter{equation}{0}

484: 

485: We consider Apollonian super-packings.

486: Analogously to part I,

487: an   Apollonian  super-packing may be considered as either

488: an geometric object or an algebraic object, as follows. \\

489: 

490: (i) [Geometric] A {\em geometric Apollonian super-packing} 

491: is a point set on the Riemann sphere

492: $\hat{\cc} = \RR^2 \cup \{ \infty\}$, consisting of 

493: all the circles in four orbits

494: of a certain group $G_{\sA^{S}}(\sD)$ of

495: M\"{o}bius transformations

496: inside  the conformal group M\"{o}b(2) acting

497: on the four circles $\{C_1, C_2, C_3, C_4\}$ in

498: a given Descartes configuration $\sD$.  The 

499: group $G_{\sA^{S}}(\sD)$ depends on $\sD$. \\

500: 

501: (ii) [Algebraic] An {\em (algebraic) Apollonian super-packing} is a

502:  set of ordered, oriented Descartes configurations,

503: given by $48$ orbits of the

504: super-Apollonian group $\sA^{S}[\sD]$. The augmented 

505: curvature-center coordinates of

506: its elements are $\sA^{S}\bW_{\sD}:=\{ \bU W_{\sD} : \bU \in \sA^{S}\}$. \\

507: 

508: 

509: 

510: A geometric super-packing  can be 

511: described  in terms of its unordered, unoriented

512: Descartes configurations. From this viewpoint,

513:  each geometric Apollonian

514: super-packing corresponds to  $48$ different algebraic 

515: super-packings; there are 

516: $24$ choices of ordering the four circles 

517: and $2$ choices of total orientation of the configuration. 

518: We can  consider geometric  super-packings as unions

519: of a countable  number of Apollonian packings. 

520: This leads to interesting questions concerning the way these

521: Apollonian packings are embedded inside the geometric super-packing.

522: We note that as  a point set, the geometric super-packing is

523: invariant under the group action $G_{\sA^{S}}(\sD)$. 

524: However it is not a closed set,   and its closure is

525: the entire Riemann sphere $\RR^2 \cup \{ \infty\}$.

526: 

527: A large part of the paper considers 

528: integrality properties of curvatures and centers

529: of some super-packings. These questions are mainly

530: studied using algebraic super-packings, although

531: we also consider questions concerning 

532: the associated geometric super-packing, such as

533: its group of symmetries under Euclidean motions.

534: 

535: In \S3  Theorem~\ref{th31} shows that each  geometric 

536: super-packing is 

537: a packing in the sense that the circles in it do not cross

538: each other transversally, as mentioned above.

539: Theorem~\ref{th32n} specifies certain subcollections of 

540: geometric super-packings which are genuine packings

541: in the sense that the interiors of the circles do not

542: overlap.

543: 

544: In \S4 we study integer super-packings,

545: in which all circles have integer curvatures.

546: Integer super-packings are classified up to

547: Euclidean motions by a 

548: single invariant, their {\em divisor} $g$,

549: which is the greatest common divisor of the

550: curvatures in any Descartes configuration in

551: the packing. Theorem~\ref{th41n} shows  that for

552: each $g \ge 1$ there is a unique such

553: integral super-packing, up to a Euclidean

554: motion. We also show in Theorem~\ref{th43n} 

555: that for each geometric Apollonian circle 

556: packing that is integral, there exists a

557: Euclidean motion taking it to one that is

558: strongly integral.

559: 

560: In \S5 we study strongly integral super-packings,

561: which are those whose curvatures are integral and whose

562: curvature$\times$center is also integral.

563: Strongly-integral super-packings are geometrically rigid:

564: Theorem~\ref{orbits} shows that for each integer $g \ge 1$ there are

565: exactly $8$ strongly integral geometric super-packings

566: which have divisor $g$. Here we do not allow the packings

567: to be moved by Euclidean motions.

568: 

569: In \S6 we study the relations between integer Apollonian packings

570: and strongly integral super-packings. 

571: Without loss of generality we restrict to

572: primitive integer super-packings, those with divisor $1$.

573: Theorem~\ref{th61} shows

574:  that each of the $8$ kinds of these has a large group of

575: internal symmetries, forming a crystallographic group.

576: For convenience we fix one of them and call it the

577: {\em standard strongly integral super-packing}; results proved for

578: it have analogues for the other seven. Theorem~\ref{th62}

579: shows that each primitive integral Descartes configuration

580: (except for the $(0,0,1,1)$ 

581: configuration) occurs in this packing with the center of its largest

582: circle being contained in the closed unit square

583: $0 \le x \le 1, ~ 0 \le y \le 1$, 

584: and the location of this circle center is unique.  

585: Theorem~\ref{th63} deduces  that the geometric 

586: standard strongly integral  super-packing contains a

587: unique copy of each primitive integral Apollonian packing, 

588: except for  the

589: $(0,0,1,1)$ packing, 

590: having the property that 

591: the center of its bounding circle lies inside the

592: closed unit square.

593: Figures~\ref{fig5}, \ref{fig6} and \ref{fig7} picture

594: the locations of all primitive integer Apollonian

595: packing with bounding circle of curvatures 6, 8 and

596: 9, respectively. The unit square is indicated by 

597: slightly darker shading in the figures. 

598: Note that in Figure~\ref{fig5} the three Apollonian packings

599: are generated by Descartes configurations with

600: curvature vectors $(-6, 7, 42, 43),~ (-6, 10, 15,19)$

601: and $(-6,11, 14, 15)$; these are root quadruples in 

602: the sense of \cite[Sec. 4]{GLMWY2}.

603: 

604: %

605: % Figure 5  (N=- 6 configurations)

606: %

607: %

608: \begin{figure}[htbp]

609: \centerline{\epsfxsize=3.0in \epsfbox{all6.ps}}

610: \caption{Integer 

611: Apollonian packings with bounding circle of curvature $6$.}~\label{fig5}

612: \end{figure}

613:  

614: %

615: % Figure 6  (N= -8  configurations)

616: %

617: %

618: \begin{figure}[htbp]

619: \centerline{\epsfxsize=3.0in \epsfbox{all8.ps}}

620: \caption{Integer Apollonian packings with bounding circle of curvature $8$.

621: }~\label{fig6}

622: \end{figure}

623:  

624: %

625: % Figure 7  (N=-9 configurations)

626: %

627: %

628: \begin{figure}[htbp]

629: \centerline{\epsfxsize=3.0in \epsfbox{all9.ps}}

630: \caption{Integer Apollonian packings with bounding circle of curvature $9$.

631: }~\label{fig7}

632: \end{figure}

633: 

634: 

635: 

636: In \S7 Theorem~\ref{th51} shows  that the super-Apollonian

637: group is a finite index normal subgroup of the 

638: group $Aut(Q_{D}, \ZZ)$ of integral automorphs of 

639: the Descartes quadratic form. The latter group can

640: be identified with an index $2$ subgroup of the

641: integer

642: Lorentz group $O(3, 1, \zz)$, and this identification

643: allows us to identify 

644: the super-Apollonian group with  a 

645: particular normal subgroup $\tilde{\sA}^{S}$

646: of index $96$ in  $O(3, 1, \zz)$, defined after

647:  Theorem~\ref{th51}.

648: 

649: In  \S8 we study  super-packings all of whose 

650: Descartes configurations are {\em super-integral} in the

651: sense that their augmented curvature-center coordinate

652: matrices $\bW_{\sD}$ are integer matrices. Theorem~\ref{th81n} 

653: shows  there

654: are exactly $14$ geometric super-packings of this kind.

655: 

656: The last section \S9 makes a few concluding remarks.

657: 

658: The Appendix considers minimal conditions to guarantee

659: that a Descartes configuration

660: is strongly integral.  Theorem~\ref{th101} shows

661: that a configuration is strongly integral if (and only

662: if) three of its four circles are strongly integral. 

663: 

664: 

665: 

666: 

667: 

668: 

669: %*****************************************************************************

670: %

671: % SECTION 3. Geometric super-packings

672: %

673: %*****************************************************************************

674: %

675: \section{Geometric Apollonian Super-Packings}

676: \setcounter{equation}{0}

677: 

678: 

679: In this section we consider properties of 

680: geometric Apollonian circle packings. We view such

681: a super-packing as a point set on the Riemann sphere

682: $\hat{\cc} = \cc \cup \{\infty\}.$  We first

683: note  that it is not a closed set.

684: It  is not hard to show that its  closure is

685: the whole Riemann sphere, 

686: Each geometric Apollonian packing  has  a group invariance

687: property under a certain group of M\"{o}bius transformations

688: which depends on the super-packing, which is 

689: the group generated by the countable set of  

690: groups of M\"{o}bius transformations that leave 

691: invariant some Apollonian packing contained in

692: the super-packing. 

693: 

694: Our object is  prove the following

695: ``packing'' property of geometric super-packings.

696: %

697: %

698: % Theorem 3.1

699: %

700: %

701: 

702: \begin{theorem}~\label{th31} 

703: A  geometric Apollonian super-packing is a circle

704: packing in the weak sense that no two circles belonging

705: to it cross each other transversally. Circles in the geometric 

706: super-packing may be nested, or tangent to each other.

707: \end{theorem}

708: 

709: 

710: Before giving the proof, we describe the 

711: nature of the geometric packing in terms of 

712: nesting of circles. We view the packing 

713: $\sA^{S}[\sD_0]$ as generated from an initial

714: (positively oriented) Descartes configuration $\sD_0$,

715: by multiplication by a finite set of generators of

716: the super-Apollonian group.

717: Each  circle in

718: the super-packing has a  well-defined 

719: {\em nesting depth} $d$ 

720: (relative to the generating configuration $\sD_0$)

721: which counts the number of circles in the packing which

722: include $C$ in their interior. Here the notion of

723: ``interior'' is defined with respect to  the initial Descartes

724: configuration $\sD_0$.

725: The Apollonian group

726: generators move ``horizontally'', leaving constant

727: the nesting depth of any circles they produce. 

728: The dual Apollonian group generators move ``vertically'',

729: by reflecting three of the circles in a configuration

730: into the interior of the fourth circle, they increase

731: the nesting depth by one. We show there is a unique

732: ``normal form'' word of minimal length in the generators

733: that produces a Descartes

734: configuration $\sD$ containing $C$.

735: The nesting depth of $C$ 

736: is exactly equal to the number $d$ of generators

737: of $\sA^{\perp}$ that appear in this normal form word.

738: The circles at nesting depth $0$

739: are those circles in the Apollonian packing generated by

740: $\sD_0$. Each circle $C$ at nesting depth $k \ge 1$ contains

741: a unique Apollonian packing, consisting  of it

742: plus all circles at depth $k+1$ nested inside it.

743: 

744: 

745: \paragraph{Proof of Theorem~\ref{th31}.} 

746: We view the geometric super-packing on the Riemann sphere

747: $\hat{\cc}= \cc \cup \{ \infty \},$ so the initial 

748: Descartes configuration $\sD_0$ 

749: consists of four circles on the sphere. In this case each

750: circle defines a spherical cap. We choose an ordering

751: and orientation of $\sD_0$ (this does not affect the geometry),

752: requiring that  $\sD_0$ have  positive (total)  orientation. 

753: 

754: 

755: Let the super-packing be $\sA^{S}[\sD_0] = \sA^{S} \bW_{\sD_0}.$

756: We  consider the effect of the super-Apollonian group generators

757: acting on the left on

758: the matrix $\bW_{\sD_0}$.

759: The Descartes

760: configurations in the super-packing are given by 

761: $\sD = \bU_m \bU_{m-1} \cdots \bU_1[\sD_0]$

762: with

763: each $\bU_k \in \{ \bS_1, \bS_2, \bS_3, \bS_4,

764: \bS_1^{\perp}, \bS_2^{\perp}, \bS_3^{\perp}, \bS_4^{\perp}\}$.

765: We consider words in the group measured by their {\em length} $m$.

766: The stage $m$ circles will consist of all new circles added

767: using products of $m$ generators. 

768: We may without loss of generality restrict to 

769: {\em normal form} words, which are those that satisfy

770: the two conditions:

771: 

772: (i) If $\bU_k = \bS_i$, then $\bU_{k+1} \ne \bS_i$

773: and $\bU_{k+1} \ne \bS_j^{\perp}$ with $j \ne i$.

774: 

775: (ii) If $\bU_k = \bS_i^{\perp}$, then  $\bU_{k+1} \ne \bS_i^{\perp}$. 

776: 

777: Equivalently, looking backwards, if $\bU_{k+1} = \bS_i^{\perp}$

778: then either $\bU_k= \bS_i$ or else $\bU_{k}=\bS_j^{\perp}$ 

779: for some $j \ne i$. A word may be put in

780: normal form by canceling adjacent equal generators, since

781: all $\bS_i^2 = (\bS_i^{\perp})^2= \bI$, and by moving towards

782: the right\footnote{That is, moving it towards

783: the beginning of the word.}

784: in the word as far as possible any generator $\bS_i^{\perp}$,

785: using the property that it commutes with all $\bS_j$ with  $j \ne i$. 

786: These operations eventually put a word in normal form,

787: without increasing its length. The operations do not change

788: the Descartes configuration $\sD$ it represents.

789: 

790: We prove the theorem by induction on the

791: number of symbols $m$ in a normal form word,

792: which we call the stage of the induction.

793: The induction hypotheses at stage $m$ are  as follows.

794: Here we  let $C_i$ refer to

795: the circle at row $i$ of the associated

796: (ordered, oriented) Descartes configuration.

797: 

798: 

799: (1) Each normal form word of length $m$ produces

800: either one or three new circles, according as

801: $\bU_{m}=\bS_i$, where it is the circle

802: $C_i$ or $\bU_{m}= \bS_j^{\perp}$, where it is

803: the three circles $C_i$ with $i \ne j$. 

804: 

805: (2) The nesting depth of any new circle produced

806: at this stage is equal to the number of occurrences of a letter

807: in $\sA^{\perp}$ in its generating normal form

808: word.

809: 

810: (3) Each new circle produced has empty interior,

811: when it first appears. 

812: 

813: In particular hypothesis (3) implies that all circles

814: produced at stage $m$ have disjoint interiors, and 

815: that each such circle  

816: contains no circles from  earlier stages in 

817: its interior. 

818: If the induction is proved, then 

819: hypothesis  (3) guarantees  that the nesting depth of a

820: circle is well-defined when it is first produced,

821: because no new circle will ever include it in its

822: interior. Hypothesis  (3) also  guarantees  that all

823: circles produced by the end of stage $m+1$ do

824: not cross. As a consequence  no two circles in

825: the packing cross, using  property (3) 

826: applied at that level $m$ which is 

827: the greater of the levels of the two circles.

828: Thus the theorem will follow.

829: 

830: The base case $m=0$ of the

831: induction is immediate, consisting of the

832: initial Descartes configuration $\sD_0$.

833: We now show the induction step for $m+1$, given $m$. 

834: We are given  a Descartes configuration

835: $\sD' = \bU_{m+1} [\sD] = \bU_{m+1} \bU_m \cdots \bU_1[\sD_0]$

836: with a normal form word. 

837: 

838: To establish hypothesis (1) for $m+1$, suppose first

839: that  $\bU_{m+1} = \bS_i^{\perp}.$

840: We assert that  the $i$-th circle $C_i$ of $\sD$ was a new

841: circle produced at stage $m$. For either $\bU_m = \bS_i$,

842: in which case it was the unique new circle in $\sD$

843: by induction hypothesis (1) at stage $m$, or $\bU_m = \bS_j^{\perp}$

844: with $j \ne i$, in which case it was one of three new circles

845: produced at stage $m$. By induction hypothesis (2) the

846: circle $C_i$ has empty interior at stage $m$. The

847: three circles $\{C_j^{'}: j \ne i\}$ in 

848: the new configuration $\sD'$  are contained

849: in the interior of  $C_i$ are therefore new circles.

850: They do not cross, being part of a Descartes configuration.

851: Thus hypothesis (1) holds for $m+1$ in this case. 

852: 

853: Suppose next that $\bU_{m+1} = \bS_i$,

854: so the possible new circle is $C_i^{'}$.

855: If 

856: $\sD_k:=\bU_k \bU_{k-1} \cdots \bU_1[\sD_0]$ 

857: is  the maximal 

858: length subword

859: such that $\bU_k= \bS_j^{\perp}$ for some $j$, 

860: with $1 \le k \le m$ then 

861: $\sD'$ belongs to the Apollonian packing generated by

862: $\sD_k$, since all subsequent generators belong to the

863: Apollonian group. If no such $k$ exists, then

864: $\sD'$ is in the Apollonian packing generated

865: by the original Descartes configuration $\sD_0$.

866: For $k \ge 1$, this  Apollonian packing is

867: entirely contained  in

868: the interior of a 

869: bounding circle $C= C_j$ first produced

870: at stage $k-1$. At that time the interior of $C$ was empty,

871: by induction hypotheses (2) and (3) at stage $k-1$.

872: The only Descartes configurations that ever can enter

873: the interior of the circle $C$, must do so by

874: a reflection in $C$ and these   are exactly

875: those normal form words

876: starting with initial segment $\bU_k \bU_{k-1} \cdots \bU_1$.

877: (This follows from uniqueness of a circle when it is

878: created, hypothesis (3) applied at stage $k$.)

879: The words contain this initial segment at the same depth,

880: with all subsequent letters in $\sA$, fill out an Apollonian

881: packing at this depth.

882:  In particular each such normal form word produces one

883: new circle in this Apollonian packing.

884: Recall that  all the circles

885: in the Apollonian packing inside the bounding circle $C$

886: have disjoint interiors (Theorem 4.1 of part I). 

887: These circles all have the same depth, and (1) holds in

888: this case. Normal form words that have another subsequent

889: generator in $\sA^{\perp}$ confine the resulting 

890: Descartes configuration to the

891: inside  of a  single circle 

892: in the Apollonian packing $\sA[\sD_k]$ already produced,

893: and all longer words with this prefix are entirely

894: contained inside this circle. In particular,

895: they do not coincide with  $C_i^{'}$.

896: the new circle produced by the normal form word corresponding

897: to $\sD'$.

898: %(and are disjoint from its interior).

899: It follows that the circle $C_i^{'}$ is new,

900: so hypothesis (1) holds for $m+1$ in this case.

901: 

902: There remains the case where $\bU_{m+1} = \bS_i$

903: and all previous $\bU_k \in \sA$. Then 

904: $\bU_{m+1} \cdots \bU_1[\sD_0]$ belongs

905: to the Apollonian packing generated by

906: $\sD_0$, and is at depth $0$. Here $C_i^{'}$

907: is new since the generation of an 

908: Apollonian packing creates one new circle

909: at each step, and any word that contains

910: an element of $\sA^{\perp}$ moves the

911: Descartes configuration inside an older

912: circle in this packing, from which it cannot

913: escape. Thus $C_i^{'}$ is a new circle in

914: this case, and hypothesis (1) holds.

915: 

916: 

917: 

918: Hypothesis (2) holds for $m+1$ in the case $\bU_m =\bS_i^{\perp}$

919: because the three new circles produced have nesting

920: depth one greater than the nesting depth of $C_i$ at

921: the previous level, to which induction hypothesis (2) applied.

922: It also holds in the remaining case $\bU_m=\bS_i$ because the 

923: argument above showed that the nesting 

924: depth did not increase in this case.

925: 

926: Hypothesis (3) holds if $\bU_{m+1} =\bS_i^{\perp}$. Now 

927: circle $C_i$ was first created only at stage $m$,

928: and the only other possible sequences leading to

929: a Descartes configuration including 

930: this circle 

931: must start from $\sD_m$ and use a generator

932: $\bU_{m+1}= \bS_j$

933: with  $j \ne i$.  In all other cases the resulting

934: Descartes configuration includes no circle inside

935: $C_i$, so the interiors of the three new circles 

936: produced are empty at the end of stage $m+1$.

937: 

938: In the remaining case  $\bU_{m+1} =\bS_i$, we have

939: already observed that the new circle $C_i^{'}$ produced,

940: is disjoint from all other circles in the Apollonian

941: packing  $\sA[\sD_k]$ created by level $m$, 

942: which are necessarily contained  in the

943: bounding circle $C$ of $\sD_k$. 

944: As mentioned above, all Descartes configurations

945: containing a circle inside $C$ must have an

946: initial segment of their generating word, giving

947: the unique normal form word that first generates $C$,

948: at stage $k \le m$. Other depth $m+1$ words with 

949: this initial segment, and with all subsequent $\bU_j$ drawn 

950: from the Apollonian group generators 

951: produce new circles in the

952: Apollonian packing $\sA[\sD_k]$, disjoint from $C_i^{'}$.

953: Any  normal  form word

954: at depth $m+1$ with this initial segment, which 

955: contain  some generator $\bS_j^{\perp}$ after-wards,

956: produce a Descartes configuration contained inside a circle

957: of the Apollonian packing $\sA[\sD_k]$

958: different from $C$. 

959: It follows that the

960: interior of $C_i^{'}$ is empty at the end of stage

961: $m+1$, as required.

962: 

963: This completes the induction step and the proof.

964: $~~~\bsq$ \\

965: 

966: Super-packings have  some properties that

967: are genuine packing properties. The proof of Theorem~\ref{th31}

968: established in hypothesis  (3) shows that the finite set of circles at

969: stage $m$ of the construction, starting from a generating

970: Descartes configuration $\sD$,  all had disjoint interiors.

971: It also gives  the following stronger result. 

972: 

973: %

974: %

975: % Theorem 3.2

976: %

977: %

978: 

979: 

980: \begin{theorem}~\label{th32n} 

981: For a geometric Apollonian super-packing given with

982: a generating Descartes configuration $\sD$, 

983: and each $k \ge 1$,  

984: the set of all circles having  nesting depth exactly $k$ 

985: with respect to $\sD$ have pairwise  disjoint interiors.

986: These circles can be viewed as  

987: forming an infinite collection

988: of Apollonian packings, each missing one circle;

989: the missing circle is a bounding circle at depth $k-1$.

990: \end{theorem}

991: 

992: \paragraph{Proof.} 

993: The proof of Theorem~\ref{th31}

994: shows that the nesting depth of a

995: circle is well-defined.  Circles at nesting depth $k$ have

996: disjoint interiors, since no two of them cross, and the

997: only way for two of them to have an interior point in

998: common is for one to be nested inside the other, which 

999: would violate the nesting ordering. 

1000: 

1001: Given a normal form word $\bU := \bU_m \bU_{m-1} \cdots \bU_1$

1002: with $\bU_m = \bS_i^{\perp}$, and containing exactly

1003: $k$ elements of 

1004: $\sA^{\perp} = 

1005: \langle \bS_1^{\perp},\bS_2^{\perp}, \bS_3^{\perp}, \bS_4^{\perp} \rangle, $

1006: the set of all normal form words having $\bV$ as prefix and all

1007: other letters $\bU_j \in \sA$ for $j > m$ produce all the

1008: circles in an Apollonian packing $\sA[\sD_m]$, 

1009: with $\sD_m = \bU[\sD_0]$.

1010: All these circles are  at depth $k$ 

1011: except for  the outer circle of $\sD_m$, which 

1012: is its $i$-th circle.

1013: Enumerating all possible such $\bU$ as prefixes  represents

1014: the set of nesting depth $k$ circles as a collection of

1015: Apollonian packings, each excluding one circle when $k \ge 1$.

1016: $~~~\bsq$ \\

1017: 

1018: \paragraph{Remarks.}

1019: (1) The proof of Theorem~\ref{th31} is a geometric analogue of the

1020: presentation for the super-Apollonian group

1021: proved in part I, \cite[Theorem 6.1]{GLMWY11}.

1022: 

1023: (2) The analogous result to Theorem~\ref{th31}

1024: fails 

1025: to hold in all dimensions $n \ge 4$, as explained

1026: in part III. The ``nesting'' property

1027: of the dual Apollonian configurations resulting

1028: from ``vertical'' moves still exists

1029: and works in all dimensions.

1030: However the ``horizontal'' motions  moving spheres  around

1031: in Apollonian packings produces  spheres that cross

1032: in dimensions $n \ge 4$, see \cite[Lemma 4.1]{GLMWY13}.

1033: 

1034: 

1035: (3) We cannot easily visualize a geometric super-packing as

1036: a completed object because the circles in it are dense

1037: in the plane. We can however picture a partial version

1038: of it that pictures all circles of size above a given

1039: threshold, in some finite region of the packing.

1040: The integral super-packings we are most interested

1041: in have a periodic lattice of symmetries (see Theorem~\ref{th61}),

1042: so it suffices to examine a finite region of the packing.

1043: Figure~\ref{fig4} in \S1 and Figure~\ref{fig8} in \S6

1044: exhibit part of a super-packing. 

1045: 

1046: (4) Every circle $C$ in a geometric super-packing $\sA^{S}[\sD]$ 

1047: has associated

1048: to it a unique Apollonian packing of which it is the bounding

1049: circle. If  it is a depth $k$ circle (relative  to the

1050: starting configuration $\sD$), then this Apollonian packing 

1051: consists of all depth $k+1$ circles contained in the interior of

1052: $C$. 

1053: 

1054: 

1055: %*****************************************************************************

1056: %

1057: % SECTION 4. Integral  Super-Packings

1058: %

1059: %*****************************************************************************

1060: 

1061: \section{Integral Super-Packings}

1062: \setcounter{equation}{0} 

1063: 

1064: An Apollonian super-packing is {\em integral} if it

1065: contains one (and hence all) Descartes configuration

1066: whose circles have integer curvatures.

1067: 

1068: An invariant of an integral super-packing is its

1069: {\em divisor} $g$, which is the greatest common divisor

1070: of the curvatures of the circles in any Descartes

1071: configuration in the super-packing.

1072: The quantity  $g$ is well-defined   independent of

1073: the Descartes configuration chosen, using the

1074: relation $\bM_{\sD'} = \bU \bM_{\sD}$ between two

1075: such configurations, where $\bU \in \sA^{S}$. Since $\bU$

1076: is an integer matrix with determinant $\pm 1$, it preserves

1077: the greatest common divisor of each column of the integer

1078: matrix $\bM_{\sD}$. Here  the first column

1079: encodes the curvatures of the circle. 

1080: 

1081: %

1082: %

1083: % Theorem 4.1

1084: %

1085: %

1086: 

1087: \begin{theorem}~\label{th41n} 

1088: For each integer $g \ge 1$ there exists an

1089: integral Apollonian super-packing with 

1090: divisor $g$. The associated geometric super-packing

1091: is unique, up to a Euclidean motion.

1092: \end{theorem}

1093: 

1094: As an immediate corollary of this result, the

1095: integral super-packing with divisor $g$ contains

1096: at least one  copy of every integral Descartes configuration

1097: having divisor $g$. For each such a Descartes configuration

1098: generates an integral super-packing with divisor $g$,

1099: so the corollary follows by the uniqueness assertion.

1100: 

1101: We defer the proof of 

1102: Theorem \ref{th41n} to the end of the section.

1103: It is based on a reduction theory which finds inside

1104: any such integral super-packing a Descartes configuration

1105: having particularly simple curvatures. 

1106: %

1107: %

1108: % Theorem 4.2

1109: %

1110: %

1111: 

1112: \begin{theorem}~\label{reduction}

1113: Let $\sD$ be an integral Descartes configuration, with

1114: divisor $g := \gcd(b_1, b_2, b_3, b_4)$.

1115: Then the Apollonian super-packing $\sA^{S} [\sD]$

1116: generated by $\sD$ contains a Descartes configuration having

1117: curvature vector a permutation of either $(0,0, g, g)$

1118: or $(0,0, -g,-g)$, with the former case occurring if

1119: $b_1+b_2+b_3+b_4 > 0$ and the latter case if $b_1+b_2+b_3+b_4 < 0$. 

1120: \end{theorem}

1121: 

1122: \noindent {\bf Proof.} 

1123: Since the super-Apollonian group $\sA^S$ preserves the (total) orientation 

1124: of Descartes configurations, it is sufficient to show that for a positively 

1125: oriented integral Descartes configuration $\sD$ with curvatures

1126: $(b_1, b_2, b_3, b_4)$, $b_1+b_2+b_3+b_4 > 0$, there exists $\bU \in \sA^S$ 

1127: and a permutation matrix $\bP_\sigma$ such that

1128: $$\bP_\sigma \bU(b_1, b_2, b_3, b_4)^T=(0,0,g,g)^T.$$

1129: 

1130: 

1131: We measure

1132: the {\em size} of the curvature vector $\bv = (b_1, b_2, b_3, b_4)^T$

1133: of a  Descartes configuration  by

1134: $$ size(\bv ):= \bo^T \bv = b_1 + b_2 + b_3 + b_4.$$

1135: We claim that for positively oriented integral Descartes configurations

1136: with greatest common divisor $g$, we have

1137: $$size(\bv)\geq 2g,$$

1138: and equality holds if and only if $\bv$ is a permutation of $(0,0,g,g)$.

1139: If all curvatures are nonnegative this is clear, since at most two

1140: can be zero, and the other two are positive integers.

1141: Now in any Descartes configuration  at most one circle can have

1142: negative curvature, call it $b_1=-a$  ($a \in \ZZ_+$), in which case  

1143: it encloses the other three.

1144: Each of these three enclosed circles has a larger curvature in

1145: absolute value than

1146: the bounding circle, so  $b_i \geq a + 1$ for $i=2,3,4$. Thus

1147: $size(\bv) \geq -a+ 3(a + 1) \geq 2a+3 > 2g,$

1148: which proves the claim.

1149: 

1150: We give a reduction procedure which

1151: chooses  matrices in $ \sA^{S}$

1152: to reduce the size and show that the procedure halts only at a

1153: vector of form $(0, 0, g,g)$, up to a permutation.

1154: To specify it, we observe that for the curvature vector 

1155: $\bv=(b_1, b_2, b_3, b_4)^T$ of 

1156: any integral

1157: Descartes configuration  with

1158: $b_1 \leq b_2\leq b_3\leq b_4$, we have

1159: \beql{605aa}

1160: size(\bS_4 \bv) = \bo^T \bS_4\bv \leq \bo^T \bv = size(\bv).

1161: \eeq

1162: and equality holds

1163: if and only if $b_1b_2+b_2b_3+b_3b_1=0$. To see this,

1164: we have $\bS_4\bv=(b_1, b_2, b_3, b_4')^T$ where

1165: $$b_4'=2(b_1+b_2+b_3)-b_4 =b_1+b_2+b_3-2\sqrt{b_1b_2+b_2b_3+b_3b_1}.$$

1166: Thus

1167: $$\bo^T \bS_4 \bv-\bo^T \bv=b_4'-b_4=-4\sqrt{b_1b_2+b_2b_3+b_3b_1} \leq 0.$$

1168: Equality can hold if and only if $b_1b_2+b_2b_3+b_3b_1=0$, which proves the

1169: observation. Note that $b_1 \leq 0$ when the equality in \eqn{605aa} holds.

1170: Also note that $g=\gcd(b_1, b_2, b_3, b_4)$ is an invariant under the 

1171: action of $\sA^S$. 

1172: 

1173: Starting with any  positively oriented integral Descartes configuration with

1174: curvature vector $(b_1, b_2, b_3, b_4)^T$, 

1175: where $b_i$ is the largest number, we apply

1176: $\bS_i \in \sA$.

1177: By \eqn{605aa},  the $size(\bv)$ decreases but cannot be negative,

1178: so after a finite

1179: series of $\bS_i$ we arrive at a positively oriented 

1180: integral Descartes configuration with curvatures 

1181: $\bv'= (b_1', b_2', b_3', b_4')^T$,

1182: where $\gcd(b_1', b_2', b_3', b_4')=g$ and

1183: the smallest curvature, say $b_1'$, satisfies $b_1' \leq 0$, and 

1184: the size of $\bv'$ can not be reduced by the action of the Apollonian group

1185: $\sA$.

1186: Call this the basic reduction step. Note that the basic reduction step

1187: involves only matrices in the Apollonian group and therefore

1188: moves around inside a single Apollonian packing.

1189: 

1190: If $b_1' =0$ then necessarily $b_2'=0$,

1191: whence the curvature vector is $(0,0,b_3',b_3')^T$,

1192: and by $g=\gcd(b_1',b_2', b_3', b_4')$, we have 

1193: $b_3' = g$ and the reduction halts.

1194: If $b_1'<0$, applying $\bS_1^T$, we get a new Descartes configuration with

1195: $\bv''= (-b_1', b_2'+2b_1', b_3'+2b_1', b_4'+2b_1')^T$,

1196: which is positively oriented and lies in a new Apollonian packing

1197: and has $$size(\bv'') = size(\bv) + 4 b_1' < size(\bv).$$

1198: Thus the size strictly

1199: decreases and is non-negative. Now we may re-apply the basic reduction step.

1200: Continuing in this way we get strict decrease of $size(\bv)$ at each

1201: step, with the only possible halting step being the smallest curvature 

1202: equals $0$. Since the size of the curvature vector 

1203: is bounded below and decreases by at least one at each

1204: step, the procedure terminates at $(0,0,g,g)$, up to a permutation.~~~$\bsq$

1205: 

1206: 

1207: \paragraph{Proof of Theorem~\ref{th41n}.}

1208: For existence, the  super-packing generated by

1209: a Descartes configuration with curvature vector $(0,0, g,g)$,

1210: which is a homothetically scaled version of the configuration

1211: $(0,0,1,1)$ pictured in Figure 1, is necessarily integral 

1212: with divisor $g$.

1213: 

1214: For uniqueness,

1215: Theorem~\ref{reduction} shows that any two geometric integral Apollonian

1216: super-packings with divisor $g$ each contain a Descartes configuration

1217: whose curvatures are $(0,0,g,g)$ up to permutation and orientation. 

1218: Now  it is true for any two such Descartes configurations with identical

1219: curvature vectors are congruent,

1220: i.e. one is obtainable from the other by a Euclidean motion.

1221: This is obvious by inspection for the $(0,0,g,g)$ Descartes

1222: configuration, which necessarily

1223: consists  of two touching circles of radius $\frac{1}{g}$

1224: and two parallel lines.

1225: 

1226: Now the Euclidean motion that takes one

1227: Descartes configuration to the other, also takes the super-packing

1228: generated by the first configuration  to

1229: the one generated by the  other, because the super-packing is defined by the

1230: action of the super-Apollonian group on the left on the

1231: Descartes configuration $\bW_{\sD}$, and this commutes with

1232: the Euclidean motion acting as a M\"{o}bius transformation

1233: on the right. This establishes uniqueness.

1234: ~~~$\bsq$

1235: 

1236: We can use the freedom of

1237: a Euclidean motion allowed in Theorem~\ref{th41n}

1238: to make an internal Apollonian super-packing strongly integral.

1239: 

1240: \begin{theorem}~\label{th43n}

1241: For each integral geometric Apollonian super-packing 

1242: there is a Euclidean

1243: motion that takes it to a strongly integral geometric

1244: Apollonian super-packing.

1245: \end{theorem}

1246: 

1247: \paragraph{Proof.}

1248: Using Theorem~\ref{reduction}  

1249: each integral geometric Apollonian packing contains 

1250: a Descartes configuration $\sD$ with curvatures $(0,0, g,g)$; 

1251: note that for a  geometric packing the order and orientation of the

1252: Descartes configuration do not matter.  The curvature vector

1253: determines the Descartes configuration up to 

1254: congruence. We can now find a strongly integral

1255: Descartes configuration $\sD'$ with this curvature vector.

1256: For  $g=1$ such a configuration is given explicitly by

1257: \eqn{N601} below, and for larger $g$ we obtain a strongly

1258: integral configuration from it using the homothety 

1259: $(x, y) \mapsto \frac{1}{g}(x, y)$. There exists a Euclidean

1260: motion that maps $\sD$ to $\sD'$, since they are congruent

1261: configurations. This motion maps the super-packing $\sA^{S}[\sD]$

1262: to the super-packing $\sA^{S}[\sD']$,  which is strongly integral.

1263: ~~~$\bsq$

1264: 

1265: %*****************************************************************************

1266: %

1267: % SECTION 5. 

1268: %

1269: %*****************************************************************************

1270: \section{Strongly Integral Super-packings} 

1271: \setcounter{equation}{0}

1272: 

1273: A Descartes configuration $\sD$ is {\em strongly integral}

1274: if its associated $4 \times 3$

1275: curvature center-coordinate matrix 

1276: $\bM_{\sD}$ is an integer matrix; this property is

1277: independent of ordering or orientation of the

1278: Descartes configuration.

1279: A super-packing is called {\em strongly integral} if

1280: it contains one (and hence all) Descartes configurations 

1281: having 

1282: this property. Since a strongly integral super-packing

1283: is integral, it has a divisor $g$ as an invariant.

1284: 

1285: Our main object in this section is to classify strongly

1286: integral super-packings, as follows.

1287: 

1288: %

1289: %

1290: % theorem 5.1

1291: %

1292: %

1293: \begin{theorem}~\label{orbits}

1294: (1) For each $g \ge 1$ there  are exactly $8$ 

1295: different geometric Apollonian

1296: super-packings that are strongly integral

1297: and have divisor $g$.

1298: 

1299: (2) The set of all ordered, oriented Descartes configurations

1300: that are strongly integral and have a given divisor $g$

1301: fill exactly 384 orbits of the super-Apollonian group.

1302: \end{theorem}

1303: 

1304: This theorem classifies  these super-packings as rigid objects, not

1305: allowed to be moved by Euclidean motions. 

1306: To prove this result we derive a normal form for

1307: a ``super-root quadruple'' in a super-packing of the kind

1308: above, as follows. 

1309: 

1310: %

1311: % theorem 5.2

1312: %

1313: %

1314: \begin{theorem}\label{normal_form} 

1315: Given a strongly integral Apollonian super-packing $\sA^s[\sD_0]$ with the 

1316:  divisor  $g \geq 1$, 

1317: there exists a unique ``reduced'' Descartes configuration 

1318: $\sD \in \sA^S[\sD_0]$ 

1319: whose curvature-center coordinate matrix $\bM=\bM_\sD$ is of the form

1320: $\bA_{m,n}[g]$ or $\bB_{m,n}[g]$ for  $m, n \in \{0, 1\}$,  

1321: up to a permutation of rows,  where

1322: \beq \label{ABmn} 

1323: \bA_{m,n}[g]=\pm \left[ \begin{array}{ccc} 

1324:      0 & 0 & 1 \\

1325:      0 & 0 & -1 \\

1326:      g & m & n \\

1327:      g & m-2 & n \end{array}  \right]

1328: \qquad  

1329: \bB_{m,n}[g]=\pm \left[ \begin{array}{ccc} 

1330:      0 & 1 & 0 \\

1331:      0 & -1 & 0 \\

1332:      g & m & n \\

1333:      g & m & n-2 \end{array}  \right],

1334: \eeq

1335: and the sign is determined by the orientation of $\sD_0$. 

1336: \end{theorem}

1337: \pf For a strongly integral Apollonian super-packing $\sA^S[\sD_0]$ with the 

1338: divisor  $g$, 

1339: by Theorem \ref{reduction} there exists a

1340: strongly integral Descartes configuration $\sD \in \sA_S[\sD_0]$ with

1341: curvatures $\pm(0,0,g,g)$. The two 

1342: straight lines in $\sD$ must be parallel to either $x$-axis or $y$-axis.

1343: It follows that the $4\times 3$ curvature-center coordinate matrix $\bM_\sD$

1344: is of the form $\bA_{m,n}[g]$, or $\bB_{m,n}[g]$, for some $m, n \in \ZZ$, 

1345: up to a permutation of rows. 

1346: 

1347: We now reduce $m, n$ to take the values $0, 1$ using the

1348: following identities, which are easy to check.

1349: \begin{eqnarray*}

1350: \bP_{(34)}\bS_3 \bA_{m,n}[g]=\bA_{m-2, n}[g],  & \qquad&

1351: \bP_{(34)}\bS_3 \bB_{m,n}[g]=\bB_{m, n-2}[g], \\

1352: \bP_{(34)}\bS_4 \bA_{m,n}[g]=\bA_{m+2, n}[g], & \qquad &

1353: \bP_{(34)}\bS_4 \bB_{m,n}[g]=\bB_{m, n+2}[g], \\

1354: \bP_{(12)}\bS_1^T \bA_{m,n}[g]=\bA_{m,n+2}[g], & \qquad&

1355: \bP_{(12)}\bS_1^T \bB_{m,n}[g]=\bB_{m+2,n}[g], \\

1356: \bP_{(12)}\bS_2^T \bA_{m, n}[g]=\bA_{m, n-2}[g],  & \qquad &

1357: \bP_{(12)}\bS_2^T \bB_{m, n}[g]=\bB_{m-2, n}[g].

1358: \end{eqnarray*}

1359: where $\bP_{(ij)}$ is the permutation matrix that exchanges $i$ and $j$. 

1360: Also note that  $\bP_\sigma \bS_i=\bS_{\sigma(i)} \bP_\sigma$,

1361: $\bP_\sigma \bS_i^T =\bS_{\sigma(i)}^T \bP_\sigma$.  Hence there is

1362: a series of group operations in $\sA^S$  which takes $\bM_\sD$ to 

1363: a permutation of $\bA_{m,n}[g]$ or $\bB_{m,n}[g]$ with $m, n \in \{0, 1\}$. 

1364: This proves the existence of the ``reduced'' Descartes configuration 

1365: in the Apollonian super-packing $\sA^S[\sD_0]$. 

1366: 

1367: To prove the uniqueness, it suffices to show that the 

1368: $24 \times 8 \times 2=384$ 

1369: Descartes configurations whose curvature-center coordinate matrices

1370: are 

1371: $$

1372: \{ \bP \bA_{m,n}[g], \bP \bB_{m,n}[g] \ | \ \bP \in \text{Perm}_4, m, n \in \{0, 1\}

1373: \}

1374: $$ 

1375: are in different Apollonian super-packings. 

1376: (There are two signs for each of $\bA_{m,n}[g]$ and $\bB_{m,n}[g]$. )

1377: In what follows we let $\tilde{\bA}_{m,n}[g]$, and $\tilde{\bB}_{m,n}[g]$

1378: denote the unique $4 \times 4$ augmented curvature-center coordinate

1379: matrices extending $\bA_{m,n}[g]$ and $\bB_{m,n}[g]$, respectively;

1380: uniqueness holds  by Theorem 3.1 of part I.

1381: 

1382: Note that each $\bS_i$ and $\bS^T_i$ preserves the (total) orientation 

1383: of the Descartes configuration, as well as the parity of every 

1384: element of $\bM_\sD$. 

1385: First we show that the matrices 

1386: $\bA_{m,n}[g], \bB_{m,n}[g], (m,n \in \{0,1\}$ 

1387: are in distinct orbits of $\sA^S \times \text{Perm}_4$.

1388: To see this, for each integral vector $\bv \in \zz^4$, 

1389: let $\kappa(\bv)$ be the

1390: number of even terms in $\bv$, and

1391: for any strongly integral Descartes configuration $\sD$ 

1392: let 

1393: $$

1394: \kappa(\bM_\sD)=(\kappa(\bv_1), \kappa(\bv_2), \kappa(\bv_3)),

1395: $$

1396: where 

1397: $\bv_1, \bv_2, \bv_3$ are the column vectors of $\bM_\sD$.

1398: Then $\kappa(\bM_\sD)$ is invariant under the action of 

1399: $\sA^S \times \text{Perm}_4$.  For $m, n \in \{0, 1 \}$,

1400: $\kappa(\bA_{m,n}[g]), \kappa(\bB_{m,n}[g])$ are all distinct except

1401: $\kappa(\bA_{1, 0}[g])=\kappa(\bB_{0, 1}[g])=(*, 2, 2)$, where

1402: $*$ is $2$ if $g$ is odd, and $4$ if $g$ is even. However, 

1403: $\bA_{1,0}[g]$ and $\bB_{0,1}[g]$ can not be equivalent under the action of 

1404: $\sA^S \times \text{Perm}_4$. Arguing by contradiction, 

1405: assume that there exists a

1406: matrix $\bU \in \sA^S \times \text{Perm}_4$ such that

1407: $\bU \bA_{1, 0}[g]=\bB_{0, 1}[g]$.  This relation lifts

1408: to augmented curvature-center coordinates:  

1409: $\bU \tilde{\bA}_{1,0}[g]=\tilde{\bB}_{0,1}[g]$. 

1410: It follows that 

1411: $\bU=(\tilde \bA_{1,0}[g])^{-1} \tilde \bB_{0,1}[g]$ is unique.

1412: We can directly verify that for $m,n \in \ZZ$ 

1413: 

1414: \beq \label{aug-ABmn} 

1415: \tilde \bA_{m,n}[g]=\left[\begin{array}{cccc}

1416: 2(n+1)/g &  0 & 0 &  1  \\

1417: 2(1-n)/g &  0 & 0 & -1 \\

1418: (m^2+n^2-1)/g &   g & m & n \\

1419: ((m-2)^2+n^2-1)/g &  g & m-2 & n \end{array}\right],

1420: \nonumber \\

1421: \tilde \bB_{m,n}[g]=\left[\begin{array}{cccc}

1422: 2(m+1)/g &   0 & 1 & 0 \\

1423: 2(1-m)/g &   0 & -1 & 0 \\

1424: (m^2+n^2-1)/g &   g & m & n \\

1425: (m^2+(n-2)^2-1)/g  &   g & m & n-2 \end{array}\right], 

1426: \eeq

1427: by checking that these satisfy the identity of Theorem 3.2 of

1428: part I necessary and sufficient 

1429: to be augmented curvature-center coordinates.

1430: Now it is easy to verify that 

1431:  $\bP_{(14)}\bP_{(23)}\bD \tilde \bA_{1, 0}[g]=\tilde \bB_{0, 1}[g]$, 

1432: where $\bD=-\bQ_D$ is defined as in \S3 of Part I \cite{GLMWY11}, 

1433: \beq \label{matrixD}

1434: \bD=\frac{1}{2} \left[ \begin{array}{rrrr} 

1435:  -1 & 1 & 1 & 1  \\

1436:  1 & -1 & 1 & 1  \\

1437:  1 & 1 & -1 & 1 \\

1438:  1 & 1 & 1 & -1 \end{array} \right].

1439: \eeq 

1440: By the uniqueness 

1441: $\bP_{(14)}\bP_{(23)} \bD =\bU \in \sA^S \times \text{Perm}_4$, 

1442: which is  impossible since $\sA^S \times \text{Perm}_4$ consists of 

1443: integral matrices only, while $\bD$ has half integers.

1444: In conclusion, $\bA_{m, n}[g], \bB_{m, n}[g]$,

1445: ($m, n \in  \{0, 1\}$) are in distinct orbits of

1446: $\sA^S \times \text{Perm}_4$. 

1447: 

1448: The final step

1449: is to show that for any permutation $\bP \neq \bI$, $\bP \bA_{m,n}[g]$

1450: (resp. $\bP \bB_{m,n}[g]$) 

1451: can not be obtained from $\bA_{m,n}[g]$ (resp. $\bB_{m,n}[g]$) 

1452: by an action of $\sA^S$. That is, we claim: 

1453:  if for  a permutation matrix $\bP \in \text{Perm}_4$,

1454: there exists a matrix $\bU \in \sA^S$ such that

1455: \[

1456: \bU \bA_{m,n}[g] =\bP \bA_{m,n}[g], \qquad \text{or } \bU \bB_{m,n}[g] =\bP \bB_{m,n}[g],

1457: \]

1458: then $\bP=\bI$.

1459: 

1460: To establish the claim, 

1461: consider  again the $4 \times 4$ ACC-coordinate matrices $\tilde \bA_{m,n}[g]$ 

1462: and $\tilde \bB_{m,n}[g]$. From \S3.1 of Part I \cite{GLMWY11},  

1463: for any Descartes configuration $\sD$, 

1464: the curvature-center coordinate matrix $\bM_\sD$  

1465: can be uniquely extended to a $4 \times 4$ ACC-coordinate matrix

1466: $\bW_\sD$. It follows that  

1467: if $\bU \bA_{m,n}[g] =\bP \bA_{m,n}[g]$, then

1468: the equality holds for their $4\times 4$ ACC-coordinate matrices, i.e., 

1469: $\bU \tilde \bA_{m,n}[g] =\bP \tilde \bA_{m,n}[g]$.  It implies  

1470:  $\bU=\bP \in \sA^S \cap \text{Perm}_4$. 

1471: However, comparing the size of $\bU$ and $\bP$, where the size of 

1472: a matrix $\bU$ is defined as $f(\bU):= \bo^T \bU \bo$, we have

1473: $f(\bP):={\bf 1}^T \bP {\bf 1}=4$ for $\bP \in \text{Perm}_4$,

1474: and $f(\bU):={\bf 1}^T \bU {\bf 1} \geq 8$ for any $\bU \in \sA^S, 

1475: \bU\neq \bI$,

1476: (c.f.\S5 of Part I \cite{GLMWY11}).  Therefore 

1477: the only possibility is $\bU=\bP=\bI$. The same argument applies to 

1478: $\bB_{m,n}[g]$, and the claim follows. 

1479: 

1480: We conclude that a reduced Descartes

1481: configuration of the

1482: form \eqn{aug-ABmn} in any 

1483: strongly integral Apollonian super-packing 

1484: exists and is unique. ~~~\bsq 

1485: 

1486: 

1487: 

1488: \paragraph{Proof of Theorem~\ref{orbits}.}

1489: (1) Since there are 48 orbits of ordered, oriented

1490: Descartes configurations corresponding to each

1491: geometric super-packing, to show there are exactly

1492: $8$ geometric super-packings it suffices to  show 

1493: that the strongly integral Descartes configurations

1494: form  $384$ orbits of the Apollonian group,

1495: which we do below.

1496: 

1497:  

1498: (2) We enumerate  the complete set of 

1499: ordered, oriented Descartes configurations that

1500: are strongly integral, with greatest common divisor $g$,

1501: as follows. 

1502: By Theorem \ref{normal_form},

1503: any such Descartes configuration  

1504: is equivalent under the action of $\sA^S$ to a permutation of a Descartes 

1505: configuration whose $4\times 3$ curvature-center coordinate matrix is of 

1506: the form $A_{m,n}$ or $B_{m,n}$, with $m,n \in \{0, 1\}$. 

1507: The uniqueness of Theorem \ref{normal_form} asserts that the $24$ permutations

1508: of $A_{m,n}[g]$ ($B_{m,n}[g]$) are all in distinct orbits of the 

1509: super-Apollonian

1510: group. Considering the two choices of orientations, we get 

1511: $24 \times 8 \times 2=384$ orbits. $~~~\bsq$ \\

1512: 

1513: 

1514: 

1515: %*****************************************************************************

1516: %

1517: % SECTION 6

1518: %

1519: %****************************************************************************

1520: 

1521: \section{Primitive Strongly Integral Super-packings}

1522: \setcounter{equation}{0}

1523: 

1524: The strongly integral

1525: superpackings having  a given curvature g.c.d. $g$ are each obtainable

1526: from one  with $g=1$ by homothety.

1527: A homothety $(x,y) \mapsto  r(x,y)$ changes all curvatures

1528: by $\frac{1}{r}$ while leaving (curvature)$\times$(center)

1529: unchanged. Applying the homothety with $r = \frac{1}{g}$ 

1530: takes a strongly integral super-packing with g.c.d. $g$ to

1531: one with g.c.d. equal to $1$.

1532: 

1533: We now  consider strongly integral super-packings

1534: having $g=1$, which we call {\em primitive} super-packings.

1535: Results for them carry over easily to those with divisor $g> 1$ by 

1536: applying a homothety. Theorem~\ref{orbits} showed there are

1537: exactly $8$ such packings. 

1538: 

1539: For convenience we will single

1540: out a particular one of them and term it the {\em standard strongly

1541: integral super-packing}.

1542: We choose this to be the super-packing generated

1543: by the ordered, oriented Descartes configuration

1544: having 

1545: \begin{equation}~\label{N601}

1546: \bM_{\sD_1}= \left[

1547: \begin{array}{rrr}

1548:  0 & 0 & 1 \\

1549:  0 & 0 & -1 \\

1550:  1 & 1 & 0 \\

1551:  1 & -1 & 0

1552: \end{array}

1553: \right]~~~~~~\mbox{so~that}~~~~~~~~

1554: \bW_{\sD_1}= \left[

1555: \begin{array}{rrrr}

1556: 2 &  0 & 0 & 1 \\

1557: 2 &  0 & 0 & -1 \\

1558: 0 &  1 & 1 & 0 \\

1559: 0 &  1 & -1 & 0

1560: \end{array}

1561: \right].

1562: \end{equation}

1563: This corresponds to a $(0,0,1,1)$ Descartes

1564: quadruple, with the centers of the two circles

1565: lying along the $x$-axis and the circles touching

1566: at the origin $(0,0)$. The associated geometric

1567: integral super-packing is the one pictured in

1568: \S3. Results proved below for

1569: the standard  super-packing apply generally to all eight 

1570: primitive integral super-packings, using the Euclidean

1571: motions mapping between them described after the

1572: proof of Theorem~\ref{th61} below. 

1573: 

1574: We first show  that the geometric standard 

1575: strongly-integral super-packing

1576: has a large group of symmetries, which form  a 

1577: crystallographic group of the plane.

1578: %

1579: %

1580: % Theorem 6.1

1581: %

1582: %

1583: 

1584: \begin{theorem}~\label{th61}

1585: The geometric standard strongly integral super-packing is

1586: invariant under the following Euclidean motions:

1587: 

1588: (1) The lattice of translations $(x,y) \mapsto (x+2, y)$, 

1589: and $(x,y)  \mapsto (x, y+2)$,

1590: 

1591: (2) The reflections $(x,y) \mapsto (-x, y)$, and 

1592: $(x,y)  \mapsto (x, -y)$.

1593: 

1594: \noindent The crystallographic group generated by

1595: these motions is  the complete set of

1596: Euclidean motions leaving the geometric standard

1597: strongly integral super-packing invariant.

1598: \end{theorem}

1599: 

1600: \paragraph{Proof.}

1601: The key fact used is that the action of the super-Apollonian

1602: group on Descartes configurations commutes with the

1603: action of Euclidean motions acting on Descartes

1604: configurations as M\"{o}bius transformations. 

1605: This was shown in part I \cite[Theorem 3.3(4)]{GLMWY11}.

1606: 

1607: (1) There is a 

1608: Descartes configuration 

1609: corresponding to $\sD_0$ shifted by $2$ in the 

1610: $x$-direction and the $y$-direction; call them $\sD_0^{x}$

1611: and $\sD_0^{y}$, respectively. These are given by the

1612: actions of $\bS_{4}$ and  $\bS_{1}^{\perp}$, respectively.

1613: Treating the $x$-shift first, we then have 

1614: $$

1615: \sA^{S}[\sD_0] = \sA^{S}[\sD_0^{x}]= \sA^{S}[ \ft_{x}(\sD_0^{'})],

1616: $$

1617: in which  $\ft_{x}: \bz \mapsto \bz +2$ is 

1618: the Euclidean motion translation by $\bv=(2,0)$

1619: as in  Appendix A of part I, and the

1620: ordered, oriented Descartes configuration $\sD_0^{'}$ is

1621: a permutation of $\sD_0$. 

1622: Then the  geometric super-packing

1623: associated to $\sA^{S}[\sD_0^{'}]$

1624: is therefore identical with that of $\sA^{S}[\sD_0]$, and that of 

1625: $\sA^{S}[ \ft_{x}(\sD_0^{'})]$ translates it by $(2,0)$.

1626: Thus the geometric packing is invariant under this translation.

1627: The argument for translation by $(0,2)$ is similar. 

1628: 

1629: (2) This geometric Descartes  configuration $\sD_0$ 

1630: is invariant under the reflections $(x,y) \mapsto (-x, y)$, and 

1631: $(x,y)  \mapsto (x, -y)$ viewed as M\"{o}bius transformations.

1632: The effect of these transformations on the ordered,

1633: oriented Descartes configuration is to permute its rows.

1634: It follows as in (1) that the associated geometric 

1635: Apollonian super-packings  are identical.

1636: 

1637: To see that these motions generate the full group of Euclidean

1638: motions leaving the super-packing invariant, we observe

1639: that the full group acts discontinuously on the plane. 

1640: This is because the image of the $(0,0,1,1)$ configuration

1641: is either left fixed, or else it moves a distance of at least two

1642: in some direction, so that its circles do not overlap. 

1643: Thus it must be contained in a crystallographic group whose

1644: translation subgroup is given by (1) above. Now the only

1645: possibilities are to extend the group by a subgroup of

1646: the finite point group of motions leaving $(0,0)$ fixed (of order $8$) 

1647: leaving

1648: the lattice $\ZZ[(0,2), (2,0)]$ of translations invariant. 

1649: Here (2) gives an extension of order $4$. No larger

1650: extension occurs by observing that otherwise the image

1651: of the $(0,0, 1,1)$ and $(-1, 2, 2, 3)$ configurations at

1652: the origin would cross themselves.

1653: $~~~\bsq$ \\

1654: 

1655: One can now check that 

1656: the  $8$ primitive geometric strongly integral super-packings

1657: given by Theorem~\ref{orbits} are obtained from

1658: the standard strongly integral

1659: super-packing by $8$ cosets of the Euclidean motions 

1660: $(x, y) \mapsto (x+1, y)$,

1661: $(x,y) \mapsto (x, y+1)$ and $(x,y) \mapsto (y, x)$

1662: with respect to the symmetries in Theorem~\ref{th61}.

1663: They are specified by the location 

1664: and orientation of the $(0,0,1,1)$

1665: configuration.

1666: 

1667: Our next result  shows that every primitive integral

1668: Descartes configuration with no curvature zero

1669: occurs inside the geometric standard strongly integral super-packing

1670: in a specified location.

1671: 

1672: %

1673: % Theorem 6.2

1674: %

1675: %

1676:  

1677: \begin{theorem}~\label{th62}

1678: In the geometric standard strongly integral super-packing,

1679: for each (unordered) primitive integral Descartes quadruple

1680: $(a, b, c, d)$ except for $(0,0,1,1)$, 

1681: there exists a Descartes configuration

1682: having these curvatures, such that the

1683: center of its largest circle lies in

1684: the closed unit square

1685: $\{(x,y): 0 \le x \le 1,~ 0 \le y \le 1\}$.

1686: The location of the center of this largest circle is unique.

1687: \end{theorem}

1688: 

1689: \paragraph{Proof.}

1690: To establish  existence,

1691: we first  show that a Descartes configuration

1692: of the curvatures occurs somewhere inside 

1693: the standard integral super-packing. This holds

1694: because the super-packing generated by such

1695: a configuration is an integral super-packing

1696: with divisor $1$, which by Theorem~\ref{th41n} 

1697: is unique up to a Euclidean motion. Thus the

1698: standard strongly integral super-packing must

1699: contain an isometric copy of it. Now that

1700: we have such a configuration inside the packing,

1701: we can use  

1702: the translation symmetries in Theorem~\ref{th61}

1703: to move it so that its largest circle

1704: has center 

1705: inside the half-open square $\{ (x, y) : -1 \le x < 1, -1 \le y < 1\}$.

1706: If we have $-1\le x < 0$ then we apply the symmetry

1707: $(x,y) \mapsto (-x,y)$, while if $-1\le y < 0$ we

1708: apply $(x,y) \mapsto (x,-y)$, as necessary.

1709: 

1710: To establish uniqueness, we argue by contradiction.

1711: If uniqueness failed,  there would exist a Euclidean motion taking

1712: one of these Descartes configurations to the other. 

1713: Since a single Descartes configuration generates the

1714: entire super-packing, we conclude that the super-packing

1715: is left invariant under this extra Euclidean motion. 

1716: Theorem~\ref{th61} described all such automorphisms

1717: and all of them, except the identity,  

1718: map every point in the interior of the unit square 

1719: square strictly outside the square. This 

1720: contradicts the assumption that the center of

1721: the largest circle of the first configuration

1722: is mapped to that of the second, when

1723: at least one of these points is strictly  inside the

1724: unit square. In the remaining cases where both 

1725: centers lie on the boundary, one shows 

1726: they are must lie  the same

1727: boundary edge and that the automorphism leaves this

1728: edge fixed, so they are identical. 

1729: Note that this  argument shows in passing

1730: that the largest circle is unique, once $(0,0,1,1)$

1731: is excluded. 

1732: $~~~\bsq$ \\

1733: 

1734: We come now to the main result of this section,

1735: which asserts that the 

1736: geometric standard super-packing

1737: contains a copy of every integral Apollonian circle

1738: packing in a canonical way. The circles in the geometric

1739: standard super-packing can be foliated into 

1740: a union of geometric Apollonian packings.  

1741: 

1742: %

1743: % Theorem 6.3

1744: %

1745: %

1746: 

1747: \begin{theorem}~\label{th63}

1748: (1) Each circle in the standard super-packing

1749: with center inside the 

1750: half-open unit square

1751: $\{(x,y): 0 \le x < 1,~ 0 \le y < 1\}$

1752: is the exterior boundary circle of a unique 

1753: primitive integral Apollonian circle packing contained

1754: in the geometric standard integral super-packing.

1755: 

1756: (2) Every primitive integral Apollonian circle packing,

1757: except for the packing  $(0,0,1,1)$,

1758: occurs exactly once in this list. 

1759: \end{theorem}

1760: 

1761: \paragraph{Proof.}

1762: (1) Let the circle $C$ be given.

1763: Recall from the proof of Theorem~\ref{th31}

1764: that  there is a unique minimal length admissible sequence

1765: $\bU_m \bU_{m-1} \cdots \bU_1 [\sD_0]$  

1766: of generators of the super-Apollonian

1767: group that yield a Descartes configuration $\sD$ containing

1768: the given circle $C$, say in its $j$-th position.

1769: Admissibility

1770: requires that  $\bU_m = \bS_j$ or $\bS_i^{\perp}$ for some $i \ne j$.

1771: Then multiplying by $\bS_j^{\perp}$ also gives an 

1772: admissible sequence, and the Descartes configuration

1773: $$

1774: \sD':=\bS_j^{\perp}[\sD] =  \bS_j^{\perp} \bU_m \bU_{m-1} \cdots \bU_1 [\sD_0]

1775: $$

1776: consists of the circle

1777: $C$  plus three new circles nested inside the interior of $C$.

1778: This Descartes configuration $\sD'$ generates 

1779: an Apollonian packing having the circle $C$ as outer

1780: boundary, contained in the standard strongly integral super-packing. It

1781: is unique, because if there were a second Apollonian packing

1782: inside the bounding circle it would contain circles crossing

1783: those in the first packing, contradicting Theorem~\ref{th31}.

1784: 

1785: (2) Recall from \cite[Sect. 3 and 4]{GLMWY2}

1786: that inside  each integer Apollonian packing is a 

1787: positively oriented Descartes configuration

1788: whose absolute values of curvatures $(a, b, c, d)$ are minimal, which is

1789: called a {\em root quadruple}. 

1790: Theorem 4.1 of \cite{GLMWY2} showed that aside   from the root quadruple

1791: $(0,0,1,1)$ every root quadruple is of the form 

1792: $a< 0< b \le c \le d$. Root quadruples are characterized

1793: by satisfying the extra condition

1794: \beql{603root}

1795: a+ b+c  \ge d.

1796: \eeq

1797: All the circles in

1798: the resulting Apollonian packing are contained inside

1799: a {\em bounding circle} of curvature $N=|a|$, i.e., radius $\frac{1}{N}$.

1800: Theorem~\ref{th62} shows that for each root quadruple,

1801: with all curvatures nonzero 

1802: there is a matching Descartes configuration

1803: whose largest circle has center inside the unit square

1804: and this largest circle is unique. The Apollonian

1805: packing contained inside this circle by  is the integral packing with

1806: the given root quadruple, and it is unique

1807: by the result of (1).  (In some cases, like $(-1, 2,2, 3)$

1808: the root configuration is not unique, but the root quadruple

1809: and the packing itself

1810: are  always unique.) Thus every primitive integer Apollonian

1811: circle packing, except $(0,0,1,1)$, occurs exactly once bounding

1812: circles having center in  the closed unit square.

1813: $~~~\bsq$ \\

1814: 

1815: 

1816: 

1817: 

1818: 

1819: The initial part of the standard  super-packing to depth 200 

1820: inside the unit square is pictured in Figure~\ref{fig8} below.

1821: 

1822: 

1823: %

1824: % Figure 8  (superpacking to depth 200)

1825: %

1826: %

1827: \begin{figure}[htbp]

1828: \centerline{\epsfxsize=6.0in \epsfbox{0mod1.ps}}

1829: \caption{A ``deeper'' initial  part of a super-packing

1830: (square of sidelength  $1$)}~\label{fig8}

1831: \end{figure}

1832: 

1833: 

1834: One can make  further 

1835: computer experiments plotting the circles having various

1836: curvatures restricted $(\bmod~4)$ inside the unit

1837: square.  The results for circles

1838: having curvatures $1~(\bmod~2)$, 

1839: $2~(\bmod~4)$  and $0~(\bmod~4)$ and size at most 

1840: $200$ are pictured in the following three figures.

1841: 

1842: %

1843: %  Figure 9: (1 mod 2)

1844: %

1845: %

1846: 

1847: \begin{figure}[htbp]

1848: \centerline{\epsfxsize=6.0in \epsfbox{1mod2.ps}}

1849: \caption{Circles of curvature $1~(\bmod~2)$}~\label{fig9}

1850: \end{figure}

1851: 

1852: %

1853: %  Figure 10 : (2 mod 4)

1854: %

1855: %

1856: 

1857: \begin{figure}[htbp]

1858: \centerline{\epsfxsize=6.0in \epsfbox{2mod4.ps}}

1859: \caption{Circles of curvature $2~(\bmod~4)$}~\label{fig10}

1860: \end{figure}

1861: 

1862: %

1863: %  Figure 11 : (0 mod 4)

1864: %

1865: %

1866: 

1867: \begin{figure}[htbp]

1868: \centerline{\epsfxsize=6.0in \epsfbox{0mod4.ps}}

1869: \caption{Circles of curvature $0~(\bmod~4)$}~\label{fig11}

1870: \end{figure}

1871: 

1872: The figures empirically

1873: indicate that the following extra reflection symmetries occur;

1874: the line of symmetry is indicated on the figure  with

1875: a dotted line.

1876: 

1877: (a) $1~(\bmod~2)$ circles are symmetrical under reflection

1878: in the line $x= 1-y$.

1879: 

1880: (b) $2~(\bmod~4)$ circles are symmetrical under reflection

1881: in the horizontal line $y= \frac{1}{2}$.

1882: 

1883: (c) $0~(\bmod~4)$ circles are symmetrical under reflection

1884: in the vertical line $x = \frac{1}{2}$.

1885: 

1886: 

1887: Based on this experimental evidence,

1888: one  of the authors (CLM) conjectured that

1889: these symmetries hold.

1890: They  were subsequently proved by

1891: S. Northshield ~\cite{No03}.

1892: 

1893: We may also illustrate these symmetry properties at the

1894: level of circles of a fixed curvature. Recall from \S3

1895: that each such circle contains a unique Apollonian packing

1896: having it as outer circle. Such a circle

1897: can then be  labelled by  root quadruple in the

1898: sense of \cite{GLMWY2} of this integral Apollonian packing.  

1899: The cases of curvature $-6, -8,$ and $-9$ corresponding to

1900: the three cases above were pictured in \S2.

1901: In our paper considering number-theoretic

1902: properties of integral Apollonian packings, it was shown

1903: in  \cite[Theorem 4.2]{GLMWY2} that the number of distinct primitive

1904: integral Apollonian packings with a given curvature $-n$ 

1905: of the outer circle had an interpretation as a class number

1906: $h^{\pm}(-4n^2)$ of positive definite binary quadratic forms of discriminant

1907: $\Delta= -4n^2$, under $GL(2, \zz)$-equivalence. One can raise 

1908: the  question whether there is some interpretation

1909: of the extra symmetries (a)-(c) above in terms of the 

1910: associated class group structure under $SL(2, \zz)$ equivalence. 

1911: 

1912: 

1913: 

1914: 

1915: 

1916: 

1917: 

1918: %*****************************************************************************

1919: %

1920: % SECTION 7

1921: %

1922: %*****************************************************************************

1923: %

1924: \section{The Super-Apollonian Group has Finite Covolume}

1925: \setcounter{equation}{0}

1926: In this section we show that the super-Apollonian group

1927: is of finite volume as a discrete subgroup of

1928: the real Lie group 

1929: $Aut(Q_{\sD}) \simeq O(3, 1)$. This 

1930: follows from  the fact that

1931: the integer Lorentz group $O(3, 1; \ZZ)$ has

1932: finite covolume in $O(3, 1)$, and the following result.

1933: 

1934: %

1935: %

1936: % theorem 7.1

1937: %

1938: %

1939: 

1940: \begin{theorem}~\label{th51} 

1941: (1) The super-Apollonian group $\sA^{S}$ is a normal

1942: subgroup of index 48 in the group $Aut(Q_{D}, \ZZ)$.

1943: The group $Aut(Q_{D}, \ZZ)$ is generated by

1944: the super-Apollonian group and the finite

1945: group of order 48 generated by the $4 \times 4$

1946: permutation matrices and $\pm \bI$.

1947: 

1948: (2) The super-Apollonian group $\sA^{S}$is a normal subgroup

1949: of index 96 in the group

1950: $G= \bJ_0~ O(3, 1; \ZZ) \bJ_0^{-1}$, where

1951: $$

1952: \bJ_0=\frac{1}{2} \left[ \begin{array}{rrrr}

1953: 1 & 1 & 1 & 1 \\

1954: 1 & 1 & -1 & -1 \\

1955: 1 & -1 & 1 & -1 \\

1956: 1 & -1 & -1 & 1 \end{array} \right].

1957: $$

1958: The group $G$ is generated by $Aut(Q_{D}, \ZZ)$

1959: and the duality matrix $\bD$, and consists

1960: of matrices with integer and half-integer entries.

1961: \end{theorem}

1962: 

1963: The duality matrix $\bD$ is given by (\ref{matrixD}) in \S5.

1964: This result  allows us to associate to the

1965: super-Apollonian group the normal

1966: subgroup 

1967: $\tilde{\sA}^{S} := \bJ_0^{-1} \sA^{S} \bJ_0$

1968: of the integer Lorentz group $O(3, 1; \ZZ)$, of index $96$.

1969: 

1970: The proof of Theorem \ref{th51} will be

1971: derived in a series of four lemmas.  

1972: Let $\Gamma$ be the group generated by adjoining to $\sA$

1973: the elements of the finite group of order $48$ given by

1974: $\text{Perm}_4$ and $\pm I$, and let

1975: $\tilde \Gamma=< \Gamma, \bD>$. 

1976: The lemmas will  prove that $\Gamma= Aut(Q_D, \zz)$

1977: and $\tilde \Gamma=\bJ_0~ O(3, 1; \ZZ) \bJ_0^{-1}$.

1978: Then analysis of cosets of $\sA$ in these groups 

1979: permits determining the index of $\sA$ in these groups

1980: and showing normality. 

1981: 

1982: To determine $\Gamma$ and $\tilde{\Gamma}$ we will show

1983: $$

1984: \Gamma \leq Aut(Q_D, \zz) \leq \tilde{\Gamma} = G.

1985: $$

1986: It is easy to show that  

1987: $\Gamma$ is a subgroup of $\tilde \Gamma$ of index 2,

1988: and $\bD \in \tilde \Gamma$ but $\bD \notin Aut(Q_D, \zz)$,

1989: so we get  

1990: $Aut(Q_D, \zz) =\Gamma =<\sA^S, \text{Perm}_4, \pm \bI>$. 

1991: It is easy to check the

1992: inclusion  $\Gamma \leq Aut(Q_D, \zz)$ since the generators 

1993: $\bS_i, \bS^T_i$, $\bP_\sigma$ and $\pm \bI$ of $\Gamma$ are all in 

1994: $Aut(Q_D, \zz)$. The inclusion $Aut(Q_D, \zz) \leq G$ is proved in

1995: Proposition \ref{lemma52}.  The equation $G=\tilde \Gamma$ is proved 

1996: in the next three lemmas.

1997: In Lemma \ref{lemma53} we prove that the integer Lorentz group is exactly 

1998: the group $Aut(L_\zz)$ of 

1999: invertible linear transformations that leave the integral Lorentz cone 

2000: $L_\zz$ invariant.  Lemma \ref{lemma54} states that in order to show 

2001: that a group $\sG$ of invertible 

2002: linear transformations of $L_\zz$ equals $Aut(L_\zz)$, 

2003: it is enough to show that (1) the action of $G$ on $L_\zz$ is transitive, 

2004: and (2) there exists a point $v \in L_\zz$ such that the stabilizer 

2005: $\sS_v =\{ \bU \in Aut(L_\zz) \ | \ \bU v =v \}$ is a subset of $\sG$. 

2006: Using Lemma \ref{lemma54}, we check that (1) the action of 

2007: $\bJ_0 \tilde \Gamma \bJ$ on $L_\zz$ is transitive, and (2), 

2008: $\bJ_0 \tilde \Gamma \bJ$ contains the stabilizer $\sS_v$ of the point

2009: $v =(1,1,0,0) \in L_\zz$. This proves $G= \tilde \Gamma$. 

2010: 

2011: %

2012: %

2013: % Lemma 7.2

2014: %

2015: 

2016: \begin{lemma} \label{lemma52} It is true that

2017: $$Aut(Q_D, \zz) \leq G= \bJ_0 O(3,1;\zz) \bJ_0^{-1}.$$

2018: \end{lemma}

2019: \pf

2020: Let $\bU \in Aut(Q_D, \zz)$. We need to show that the entries of

2021: $\bJ_0^{-1} \bU \bJ_0 =\bJ_0 \bU \bJ_0$ are all integers. Since

2022: $\bJ_0= \frac{1}{2}\bo \bo^T -\bT$, where

2023: \[

2024: \bT=\left[ \begin{array}{cccc}

2025: 0 & 0 & 0 & 0 \\

2026: 0 & 0 & 1 & 1 \\

2027: 0 & 1 & 0 & 1 \\

2028: 0 & 1 & 1 & 0 \end{array} \right],

2029: \]

2030: we have

2031: \beq

2032: \bJ_0 \bU \bJ_0 = \frac{1}{4}(\sum_{i,j} U_{ij})\bo \bo^T -\frac{1}{2}\bo \bo^T

2033: \bU \bT -\frac{1}{2} \bT \bU \bo \bo^T+\bT \bU \bT,

2034: \eeq

2035: where $\bT \bU \bT$ is an integer matrix.

2036:                                                                                 

2037: From $\bU^T \bQ_D \bU=\bQ_D$ and $ \bQ_D=\frac{1}{2}(2\bI-\bo \bo^T)$, we have

2038: \beq ~\label{6.2-for1}

2039: \bU ^T (2\bI-\bo \bo^T) \bU=2\bI -\bo \bo^T.

2040: \eeq

2041: Denote by $\bv_i$ the $i$-th column of $\bU$,

2042: and by $size(\bv):=\bo^T \bv$ the size of a vector $\bv$.  Equating the entries

2043: of \eqref{6.2-for1} we get

2044: $2 \bv_i \cdot \bv_j -size(\bv_i)size(\bv_j) =2\delta_{ij}-1$.

2045: In particular, $size(\bv_i)$, $i=1,2,3, 4$  are  odd integers. It follows that

2046: the matrix $\frac{1}{2}\bo \bo^T \bU \bT$ is integral.

2047: 

2048: Note that $Aut(Q_D, \zz)$ is closed under transposition. This is because

2049: \beq \label{6.2for-2}

2050: & & \bU \in Aut(Q_D, \zz) \Longrightarrow \bU^T \bQ_D \bU=\bQ_D

2051:  \Longrightarrow \bU^T \bQ_D \bU \bQ_D \bU^T=\bU^T  \nonumber \\

2052: & \Longrightarrow & \bU \bQ_D \bU^T =( \bU^T \bQ_D)^{-1} \bU^T=(\bQ_D)^{-1}=\bQ_D

2053: \Longrightarrow \bU^T \in Aut(Q_D, \zz).

2054: \eeq

2055: Applying the same argument of the preceding paragraph to $\bU^T$, we

2056: then prove that the matrix $\frac{1}{2} \bT \bU \bo \bo^T$ is integral.

2057: 

2058: 

2059: Again using $2 \bv_i \cdot \bv_j -size(\bv_i)size(\bv_j) =2\delta_{ij}-1$, 

2060: summing over

2061: $i, j =1, \dots, 4$, we get

2062: \[

2063: (\sum_{i,j} U_{i,j})^2=2(\bv_1+\bv_2+\bv_3+\bv_4)\cdot 

2064: (\bv_1+\bv_2+\bv_3+\bv_4) +8.

2065: \]

2066: Note that $\bv_1+\bv_2+\bv_3+\bv_4=(size(\br_1), size(\br_2), size(\br_3), 

2067: size(\br_4))^T$

2068: where $\br_i$ is the $i$-th row of the matrix $\bU$. The sum

2069: $\sum_i (size(\br_i))^2$

2070: is a multiple of 4 since each $ size(\br_i)$ is odd. It follows that

2071: $(\sum_{i,j} U_{i,j})^2$ is a multiple of 8. Since $\sum_{i,j} U_{i,j}$

2072: is an integer, we conclude that $\sum_{i,j} U_{i,j}$ is a multiple of 4

2073: and the matrix  $\frac{1}{4}(\sum_{i,j} U_{ij})\bo \bo^T $ is integral.

2074: This proves Lemma  \ref{lemma52}.  ~~\bsq

2075: 

2076: 

2077: Clearly $\sA^S$, $\text{Perm}_4$ and $\pm \bI$ are subgroups

2078: of $Aut(Q_D, \zz)$.  Let $\Gamma$ be the group generated by $\sA^S$,

2079: $\text{Perm}_4$ and $\pm \bI$, and let 

2080: $\tilde \Gamma := \langle\Gamma, \bD \rangle$.

2081: Then $\Gamma$ is a subgroup of $\tilde \Gamma$ of index $2.$

2082:                                                                                 

2083: The {\em Lorentz light cone} is the set of points

2084: $\{ (y_0, y_1, y_2, y_3)^T \in \rr^4 ~: \ -y_0^2+y_1^2+y_2^2+y_3^2 = 0\}$.

2085:  Let $L_\zz$ be the set of integer points in the

2086: Lorentz light cone, i,e,

2087: $$L_\zz :=\{ (y_0, y_1, y_2, y_3)^T \in \zz^4 ~:~ \ -y_0^2+y_1^2+y_2^2+y_3^2 = 0\},$$

2088: and let $Aut(L_{\zz})$ be the set of linear transformations that

2089: leave $L_\zz$ invariant.

2090: %

2091: %

2092: % Lemma 7.3

2093: %

2094: 

2095: \begin{lemma}~\label{lemma53}

2096:  $Aut(L_\zz) =O(3, 1, \zz)$.

2097: \end{lemma}

2098: \pf It is clear that $O(3,1;\zz) \subseteq  Aut(L_\zz)$. 

2099: To show the other direction,

2100: let $\bU \in Aut(L_\zz)$. For any integer point $\bv \in L_\zz$,

2101: $U\bv \in L_\zz$. Therefore $(\bU \bv)^T Q_\sL (\bU \bv)=0$, i.e.,

2102: $\bv^T (\bU^T Q_{\sL} \bU) \bv=0$.

2103: It is easy to check that the only symmetric matrices $\bQ$ satisfying

2104: $\bv^T \bQ \bv=0$ for all $\bv \in L_\zz$ are of the form 

2105: $\bQ=diag[-a, a, a, a]$.

2106: Hence $\bU^T \bQ_\sL \bU=c^2 \bQ_\sL$, where $c=det(\bU)$.

2107: 

2108: Let

2109: \[

2110: \bX=\left[\begin{array}{cccc}

2111: 1 & 1 & 1 & 1 \\

2112: 1 & -1 & 0 & 0 \\

2113: 0 & 0 & 1 & 0 \\

2114: 0  & 0 & 0 & 1 \end{array} \right].

2115: \]

2116: Note that every column vector of $\bX$ is an integer point in $L_\zz$. Let

2117: $\bY=\bU \bX$, and $\bZ=\bU^{-1} \bX$. Since $\bU \in Aut(L_\zz)$,  

2118: every column of $\bY$

2119: and $\bZ$ is also an integer point in $L_\zz$.

2120: Therefore $\det(\bY)=c \det(\bX)=-2c$,

2121: and $c \det(\bZ)=\det(\bX)=-2$. For each $(y_0, y_1, y_2, y_3)^T \in L_\zz$,

2122: we have $y_0^2=y_1^2+y_2^2+y_3^2$. It follows that

2123:  either $y_0$, $y_1$, $y_2$, $y_3$ are

2124: all even, or $y_0$ and exactly one of $y_1$, $y_2$, $y_3$ are even. 

2125: In both cases,

2126: $\det(\bY)$ and $\det(\bZ)$ are even. This forces $c=\pm 1$. Hence

2127: $\bU^T \bQ_\sL \bU =\bQ_{\sL}$, i.e., $\bU \in O(3,1)$.

2128: 

2129: Since $\bU$ maps points  $(1, \pm1, 0, 0)^T$, $(1, 0, \pm1,

2130: 0)^T$, $(1, 0, 0, \pm1)^T$ to integer points, if  $(a, b, c, d)$ is a row

2131: of $\bU$, then $a\pm b$, $a\pm c$, $a\pm d \in \zz$.

2132: Thus there exist integers $a'$, $b'$, $c'$, $d'$ of the same parity

2133: such that $a=\frac{a'}{2}$, $b=\frac{b'}{2}$, $c=\frac{c'}{2}$, 

2134: $d=\frac{d'}{2}$.

2135: However, by Equation \eqref{6.2for-2}  $\bU \bQ_\sL \bU^T =\bQ_\sL$. 

2136: This implies

2137:  $-a^2+b^2+c^2+d^2=\pm1$. Therefore ${b'}^2+{c'}^2+{d'}^2\equiv {a'}^2$

2138: (mod 4). Hence $a'$, $b'$, $c'$, $d'$ must all be even, 

2139: which means  $a$, $b$, $c$, $d$

2140: are integers, and $\bU \in O(3,1; \zz)$. ~~\bsq

2141: %

2142: %

2143: % Lemma 7.4

2144: %

2145: 

2146: \begin{lemma} \label{lemma54}

2147:  Let  $\cal G$ be a group of linear transformations that preserves

2148: $L_\zz$. If the action is  transitive, and there exists

2149: a point $v \in L_\zz$ such that the stabilizer

2150: $\sS_v=\{ \bU \in Aut(L_\zz) | \bU v=v\} \subseteq {\cal G}$,

2151: then ${\cal G}=Aut(L_\zz)$.

2152: \end{lemma}

2153: \pf  Clearly $\sG \subseteq Aut(L_\zz)$.

2154: For any $\bP \in Aut(L_\zz)$, assume $\bP(v)=v'$. Since

2155: $\sG$ acts transitively on $L_\zz$, there exists $\bG \in \sG$ such

2156: that $\bG(v')=v$. That is,  $\bG \bP(v)=v$. So $\bG \bP 

2157: \in \sS_v \subseteq \sG$, and

2158: then $\bP \in \bG^{-1} \sG=\sG$. This proves $Aut(L_\zz) \subseteq

2159: \sG$.~~~\bsq

2160: 

2161: %

2162: %

2163: % Lemma 7.5

2164: %

2165: 

2166: 

2167: \begin{lemma} \label{lemma55}

2168:   $\tilde \Gamma = G= \bJ_0 O(3,1;\zz)\bJ_0^{-1}$.

2169: \end{lemma} 

2170:                 

2171: \pf

2172:     By Lemma \ref{lemma53}, it is  sufficient to prove that

2173: \beq \label{6.2for-3}

2174: \bJ_0^{-1} \tilde \Gamma \bJ_0=\bJ_0 \tilde \Gamma \bJ_0 =Aut(L_\zz).

2175: \eeq

2176: It is straightforward to check that $\bJ_0 \bS_i \bJ_0$, $\bJ_0 \bS^T_i \bJ_0$,

2177: $\bJ_0 \bP_{\sigma} \bJ_0$ and $\bJ_0 \bD \bJ_0$ are integer matrices. 

2178: In particular,

2179: \beqs

2180: \bJ_0 \bS_1 \bJ_0=\left[ \begin{array}{rrrr}

2181:           2 & -1 & -1 & -1 \\

2182:           1 &  0 & -1 & -1 \\

2183:           1 & -1 &  0 & -1 \\

2184:           1 & -1 & -1 &  0 \end{array} \right],

2185: \eeqs

2186: $\bJ_0 \bP_{34} \bJ_0 =\bP_{34}$ and $\bJ_0 \bD \bJ_0 =diag[1,-1,-1,-1]$. 

2187: Therefore

2188: $\bJ_0 \tilde \Gamma \bJ_0 \subseteq O(3,1;\ZZ) = Aut(L_\zz)$.

2189:  

2190:    For any integer point $(y_0, y_1, y_2, y_3) \in L_\zz$,

2191: $-y_0^2+y_1^2+y_2^2+y_3^2=0$ implies \\

2192: $y_0+y_1+y_2+y_3\equiv 0~ (\bmod~ 2)$.

2193: Hence $\bJ_0 (y_0, y_1, y_2, y_3)^T$ is integral. It follows that

2194: \[

2195: \bJ_0(L_\zz)=

2196: \{(a_1, a_2, a_3, a_4) \in \zz^4 ~:~ (a_1, a_2, a_3, a_4) 

2197: \text{ are curvatures of a Descartes configuration }\}.

2198: \]

2199: Then Theorem \ref{reduction}  implies that 

2200: $\bJ_0 \tilde \Gamma \bJ_0$ acts transitively

2201: on the integral Lorentz light  cone $L_\zz$, where $-\bI$ exchanges the total

2202: orientation of a point in $L_\zz$. 

2203:                                                                               

2204: We use Lemma \ref{lemma54} to prove equation \eqref{6.2for-3}

2205: with $\sG=J_0 \tilde \Gamma J_0$.

2206: Let $\bv :=(1, 1, 0,0) \in L_\zz$ and consider its stabilizer

2207: $$\sS_{\bv} :=\{ \bU \in O(3, 1;\zz) ~:~ \bU \bv=\bv, \ \bU^T \bQ_\sL \bU 

2208: = \bQ_\sL\}.$$

2209: Assume $\bU=(u_{i,j})_{i,j=1}^4$.

2210: Solving the equations

2211:       \beq

2212: \bU(1,1,0,0)^T=(1,1,0,0), \qquad

2213: \bU^T \bQ_\sL \bU= \bQ_\sL,

2214: \eeq

2215: we obtain the following linear and quadratic relations

2216: between the entries of $\bU$:

2217: \beqs

2218: u_{12}=1-u_{11},~~~~~ u_{13}&= & u_{23},~~~~~ u_{14}=u_{24}, \\

2219: u_{22}=1-u_{21},~~~~ u_{32} &= & -u_{31},~~~~ u_{42}=-u_{41},

2220: \eeqs

2221: and

2222: \beqs

2223: u_{33}^2+u_{43}^2=u_{34}^2+u_{44}^2=1,& &\ u_{33}u_{34}+u_{43}u_{44}=0, \\

2224: u_{13}=u_{31}u_{33}+u_{41}u_{43},  & &\ u_{14}=u_{31}u_{34}-u_{41}u_{44}, \\

2225: u_{21}=\frac{1}{2}(u_{31}^2+u_{41}^2).~~~

2226: \eeqs

2227:   It follows that the matrix $\bU$ can be expressed as

2228: \beq\label{form}

2229: \bU=\left( \begin{array}{cccc}

2230: 1+t & -t & gm+hn & km+ln \\

2231: t & 1-t & gm+hn & km+ln \\

2232: m & -m & g & k \\

2233: n & -n & h & l \end{array}\right),

2234: \eeq

2235: where $t=(m^2+n^2)/2$, $g^2+h^2=k^2+l^2=1$, and $gk+hl=0$.

2236: Since $g, h, k, l \in \zz$, we must have

2237:  $(g, h), (k, l) \in \{(\pm 1, 0), (0, \pm 1)

2238: \}$.

2239:                                    

2240: We can classify the matrices of the form \eqref{form} into four types,

2241: up to a possible  multiplication by $\bP_{34}$)

2242: , as follows.

2243: \beqs

2244: \text{Type I}: \ \left( \begin{array}{cccc}

2245:  1+t & -t & m & n\\

2246:  t & 1-t & m & n\\

2247:  m & -m & 1 & 0\\

2248:  n & -n & 0 & 1 \end{array}\right),

2249: \qquad

2250: \text{Type I\@I}: \ \left( \begin{array}{cccc}

2251:  1+t & -t & m & -n\\

2252:  t & 1-t & m & -n\\

2253:  m & -m & 1 & 0\\

2254:  n & -n & 0 & -1 \end{array}\right),

2255: \eeqs

2256: \beqs

2257: \text{Type  I\@I\@I}: \ \left( \begin{array}{cccc}

2258:  1+t & -t & -m & n\\

2259:  t & 1-t & -m & n\\

2260:  m & -m & -1 & 0\\

2261:  n & -n & 0 & 1 \end{array}\right),

2262: \qquad

2263: \text{Type I\@V}: \ \left( \begin{array}{cccc}

2264:  1+t & -t & -m & -n\\

2265:  t & 1-t & -m & -n\\

2266:  m & -m & -1 & 0\\

2267:  n & -n & 0 & -1 \end{array}\right),

2268: \eeqs

2269: where $t=(m^2+n^2)/2$ and $m,n, t \in \zz$.

2270:                                                                                 

2271: We denote  a matrix of type $X$ with parameters $m,n$ by $\bU(m,n; {\rm X})$.

2272: The following equations can be easily checked.

2273: \beqs

2274:  \bU (m,n; {\rm I}) \bU (k,l; {\rm I})&=& \bU (m+k, n+l; {\rm I}),\\

2275:  \bU (m,n;\text{I\@I})  \bU (k,l; {\rm I}) & = &

2276:          \bU (m+k, n-l; \text{I\@I}),\\

2277:  \bU (m,n; \text{I\@I\@I})  \bU (k,l; \text{I}) & = &

2278:         \bU (m-k, n+l; \text{I\@I\@I}),\\

2279:  \bU (m,n; \text{I\@V})  \bU (k,l; \text{I}) & = &

2280:          \bU (m-k, n-l; \text{I\@V}).

2281: \eeqs

2282: Also we have

2283: \[

2284:  \bU (1,1; {\rm I})^{-1} =\bU (-1,-1;{\rm I}), \qquad

2285:  \bU (1, -1; {\rm I})^{-1}= \bU (-1, 1; {\rm I}).

2286: \]

2287: Therefore the stabilizer $\sS_v$ is generated by $\bP_{34}$,

2288: $\bU (0,0; {\rm I})$, $\bU (0,0; \text{I\@I})$

2289: $\bU (0,0; \text{I\@I\@I})$, $\bU (0,0; \text{I\@V})$

2290: together with $\bA= \bU (1,1;{\rm I}), \bB=\bU (1, -1;{\rm  I})$.

2291:                                                                                 

2292: Note that $\bU (0,0; \text{I})=\bU (0,0; \text{I\@I})^2$,

2293: $ \bU (0,0; \text{I\@I\@I})=\bP_{34} \bU (0,0; \text{I\@I}) \bP_{34}$,

2294: and $\bU (0,0; \text{I\@V})=\bU (0,0; \text{I\@I})

2295: \bU (0,0; \text{I\@I\@I})$.

2296: Thus $\sS_v$ is generated by the matrices $\bP_{34}$, $\bA$, $\bB$, and

2297: $\bC=\bU (0,0; \text{I\@I})$.

2298: 

2299: Now  one can check that

2300: \beqs

2301: \bP_{34} & = & \bJ_0 \bP_{34} \bJ_0 ~\in \bJ_0 \tilde \Gamma \bJ_0, \\

2302: \bC & = & {\rm diag}[1,1,1,-1] =\bJ_0(-\bI \bP_{23} \bP_{14} \bD)\bJ_0 

2303: ~\in \bJ_0 \tilde \Gamma \bJ_0,  \\

2304: \bA & = & (\bJ_0 \bS_1 \bJ_0 ){\rm diag}[1,1,-1, -1]= 

2305: \bJ_0 (\bS_1 \bP_{12} \bP_{34}) \bJ_0

2306: ~\in \bJ_0 \tilde \Gamma \bJ_0,\\

2307: \bB & = & \big(\bJ_0 \bP_{34}  \bJ_0\big)  \bC \big(\bJ_0 \bS_2 \bJ_0  \big) \bC ~\in

2308:  \bJ_0 \tilde \Gamma \bJ_0.

2309: \eeqs

2310: Hence $\sS_v \subseteq  \bJ_0 \tilde \Gamma \bJ_0.$ By Lemma \ref{lemma54}

2311: $ \bJ_0 \tilde \Gamma \bJ_0=Aut(L_\zz)=O(3,1;\zz)$, or equivalently,

2312: $\tilde \Gamma =\bJ_0 O(3,1;\zz) \bJ_0^{-1}$. This finishes the proof.

2313: ~~~\bsq \\

2314: %

2315: % proof of thm. 7.1

2316: %

2317: %

2318: 

2319: \vspace{.3cm}

2320: \noindent {\bf Proof of Theorem~\ref{th51}.}

2321: Lemma \ref{lemma52} and Lemma \ref{lemma55} show that

2322: $\Gamma \leq Aut(Q_D, \zz) \leq G$. Hence $\Gamma =Aut(Q_D, \zz)$ since

2323: $\Gamma$ is a subgroup of $G$ of index 2 and

2324: the duality matrix $\bD$ is not in $Aut(Q_D, \zz)$. In other words,

2325: $Aut(Q_D, \zz)$ is generated by the super-Apollonian group $\sA^S$ and the

2326: finite group of order 48 generated by  the $4 \times 4$ permutation matrices

2327: and $\pm I$.

2328: The super-Apollonian group $\sA^S$ is a normal subgroup

2329: of $Aut(Q_D, \zz)$,

2330: since $\bP_\sigma \bS_i \bP_{\sigma^{-1}} =\bS_{\sigma(i)}$, and

2331: $\bP_\sigma \bS_i^T \bP_{\sigma^{-1}} =\bS_{\sigma(i)}^T$. The index is 48

2332: since $\sA^S \cap (\text{Perm}_4 \times \{ \pm \bI\})=\bI$, (c.f. \S5

2333: of Part I \cite{GLMWY11}).

2334:                                              

2335:                                    

2336: By Lemma \ref{lemma55}, the group $G$ is generated by $Aut(Q_D, \zz)$

2337: and $\bD$. Note that $\bD^2=I$ and $\bD \bS_i \bD =\bS^T_i$. 

2338: It follows that the super-Apollonian group is a normal  subgroup 

2339: of $G$ with index 96. ~~\bsq \\

2340: 

2341: The second part of Theorem~\ref{th51} can be rephrased

2342: as asserting that the conjugate group

2343: $\tilde{\sA}^{S} = \bJ_0 \sA \bJ_0$ is a normal subgroup of

2344: index $96$ in $O(3,1)$. Its generators 

2345: $\tilde{\bS}_j = \bJ_0 \bS_j \bJ_0^{-1}$

2346: and $\tilde{\bS}_j^{\perp} = \bJ_0 \bS_j^{\perp} \bJ_0^{-1}$

2347: are given by

2348: $$

2349: \tilde{\bS}_1 

2350: \left[ \begin{array}{rrrr}

2351: 2 & -1 & -1 &-1 \\

2352: 1 & 0 & -1 & -1 \\

2353: 1 & -1 & 0 & -1 \\

2354: 1 & -1 & -1 & 0

2355: \end{array} 

2356: \right] ~~~\mbox{and}~~

2357: \tilde{\bS}_2  = \left[ \begin{array}{rrrr}

2358: 2 & -1 & 1 &1 \\

2359: 1 & 0 & 1 & 1 \\

2360: -1 & 1 & 0 & -1 \\

2361: -1 & 1 & -1 & 0

2362: \end{array}

2363: \right] 

2364: $$

2365: and

2366: $$

2367: \tilde{\bS}_3 = \left[ \begin{array}{rrrr}

2368: 2 & 1 & -1 &1 \\

2369: -1 & 0 & 1 & -1 \\

2370: 1 & 1 & 0 & 1 \\

2371: -1 & -1 & 1 & 0

2372: \end{array} 

2373: \right] ~~~\mbox{and}~~

2374: \tilde{\bS}_4  = \left[ \begin{array}{rrrr}

2375: 2 & 1 & 1 &-1 \\

2376: -1 & 0 & -1 & 1 \\

2377: -1 & -1 & 0 & 1 \\

2378: 1 & 1 & 1 & 0

2379: \end{array}

2380: \right] 

2381: $$

2382: The generators $\tilde{\bS}_j^{\perp} = \bJ_0 \bS_j^{\perp} \bJ_0^{-1}$

2383: are given by 

2384: $$

2385: \tilde{\bS}_j^{\perp} = (\tilde{\bS}_j)^{T},

2386: $$

2387: which follows using $\bS_j^{\perp} = \bS_j^T$ 

2388: and $\bJ_0 = \bJ_0^T = \bJ_0^{-1}.$

2389: 

2390: 

2391: 

2392: %*****************************************************************************

2393: %

2394: % SECTION 8. Super-integral Configurations

2395: %

2396: %*****************************************************************************

2397: %

2398: \section{Super-Integral Super-Packings}

2399: \setcounter{equation}{0}

2400: 

2401: This section  treats the strongest form of integrality for

2402: super-packings, which is that where one

2403: (and hence all) Descartes configurations in the

2404: super-packing have 

2405: an integral augmented curvature-center coordinate

2406: matrix $\bW_{\sD}$.

2407: We say that a Descartes configuration with this

2408: property is {\em super-integral}, and the

2409: same for the induced super-Apollonian packing.

2410: The following result classifies such packings.

2411: %

2412: %

2413: % Theorem 8.1

2414: %

2415: %

2416: 

2417: \begin{theorem}~\label{th81n}

2418: (1) These are  exactly $14$ different geometric

2419: super-packings that are super-integral.

2420: 

2421: (2) The set of ordered, oriented Descartes

2422: configurations that are super-integral comprise

2423: $672$ orbits of the super-Apollonian

2424: group.

2425: \end{theorem}

2426: 

2427: These packings are classified  here as rigid objects, not movable

2428: by Euclidean motions. 

2429: To prove this result, it suffices to determine which strongly-integral

2430: configurations are super-integral.

2431: The next result classifies 

2432: the possible types of super-integral Descartes configurations,

2433: according to the allowed value of their divisors.

2434: 

2435: %

2436: %

2437: % Theorem 8.2

2438: %

2439: %

2440: 

2441: \begin{theorem}~\label{super-int}

2442: Suppose that

2443: an ordered, oriented  Descartes configuration $\sD$ in $\RR^2$ has integral 

2444: curvature-center

2445: coordinates $\bM=\bM_{\sD}$, and let $g= \gcd(a_1, a_2, a_3, a_4)$,

2446: where $(a_1, a_2, a_3, a_4)$ is its first column of signed

2447: curvatures.

2448: Then $\sD$ has integral augmented curvature-center coordinates

2449: $\bW_{\sD}$ if and only if one of the following conditions hold. 

2450: 

2451: (i) $g=1,$ or

2452: 

2453: (ii) $g= 2$, and each row of $\bM$ has 

2454: the sum of its last two entries being $1~(\bmod~2)$. 

2455: 

2456: (iii) $g=4$, and the last columns rows of $\bM$ are congruent to

2457: $$

2458: \left[ \begin{array}{cc}

2459: 1 & 0 \\

2460: 1 & 0 \\

2461: 1 & 0 \\

2462: 1 & 0 \\ \end{array} \right]  ~~~\mbox{or}~~~

2463: \left[ \begin{array}{cc}

2464: 0 & 1 \\

2465: 0 & 1 \\

2466: 0 & 1 \\

2467: 0 & 1 \\ \end{array} \right] ~~(\bmod~2).

2468: $$

2469: \end{theorem}

2470: \pf

2471: By Theorem \ref{normal_form}, there exists a matrix $\bU \in \sA^S$ and 

2472: a permutation matrix $\bP$ such that 

2473: \[

2474: \bP \bU \bM_\sD =A_{m,n}[g] \text{ or } B_{m,n}[g],

2475: \]

2476: for some $m, n \in \{0,1\}$, 

2477: where $A_{m,n}[g]$ and $B_{m,n}[g]$ are given in Formula \eqref{ABmn}, 

2478: and their

2479: corresponding augmented curvature-center coordinate matrices are

2480: $\tilde A_{m,n}[g]$, $\tilde B_{m,n}[g]$, given in Formula \eqref{aug-ABmn}. 

2481: Since each generator of $\sA^S$ preserves the super-integrality, as

2482: well as the parity of every element of $\bW_\sD$,  it follows that 

2483: $\bW_\sD$ is integral if and only if one of the following conditions

2484: holds:

2485: \begin{enumerate}

2486: \item $g=1$, or

2487: \item $g=2$, and $m^2+n^2-1 \equiv 0~(\bmod~2)$, or

2488: \item $g=4$, and the reduced form is $A_{0,1}[4]$ or $B_{1,0}[4]$, up 

2489: to a permutation of rows.  

2490: \end{enumerate}

2491: In Case 2 we need $m+n \equiv 1 (\bmod 2)$; in Case 3 the condition is

2492: equivalent to the one stated in the theorem. 

2493: ~~~$\bsq$

2494: 

2495: 

2496: \paragraph{Proof of Theorem~\ref{th81n}.} 

2497: (1) Since each geometric super-packing corresponds to 48 distinct orbits of

2498: the super-Apollonian group on ordered, oriented Descartes configurations, 

2499: to show  there are exactly 14 geometric super-packings,

2500: it suffices to show there are exactly 672 orbits of the super-Apollonian

2501: group that 

2502: are super-integral, which is (2).

2503: 

2504: 

2505: 

2506: 

2507: (2)Theorems \ref{normal_form} and \ref{super-int} allow us to classify the 

2508: set of ordered, oriented Descartes configurations that are super-integral 

2509: by the action of $\sA^S$.  

2510: From the criterion of Theorem \ref{super-int}, we have:

2511: \begin{enumerate}

2512: \item For $g=1$, any strongly integral Descartes configuration is 

2513: super-integral. 

2514: \item For $g=2$, half of those orbits are super-integral, namely, those

2515:  whose reduced forms are $A_{0,1}[2], A_{1,0}[2], B_{0,1}[2]$ or $B_{1,0}[2]$, 

2516: up to 

2517:  a permutation of rows.

2518: \item For $g=4$, one fourth of those orbits are super-integral, namely, 

2519: those whose reduced form are $A_{0,1}[4]$ or $B_{1,0}[4]$, up to a permutation 

2520: of rows. 

2521: \item For $g \neq 1, 2, 4$, there are no super-integral Descartes 

2522: configurations.   

2523: \end{enumerate}

2524: 

2525: 

2526: 

2527: \begin{table}[htpb] 

2528: \begin{center}

2529: \begin{tabular}{|c|c|c|} 

2530: \hline 

2531: $\bg$  & \# of orbits of $\sA^S$  & Representative \\

2532: \hline

2533: $(1,1,1,1)$   & 96   & $A_{1,1}[1],~~~ B_{1,1}[1]$ \\

2534: $(2,1,1,1)$   & 96   & $A_{1,0}[1], ~~~B_{0,1}[1]$ \\

2535: $(1,1,2,1)$   & 48   & $A_{0,0}[1]$              \\ 

2536: $(1,1,1,2)$   & 48   & $B_{0,0}[1]$  \\

2537: $(4,1,2,1)$    & 48   & $A_{0,1}[1]$  \\

2538: $(4,1,1,2)$    & 48   & $B_{1,0}[1]$  \\ \hline 

2539: $(1,2,1,1)$    & 96   & $A_{1,0}[2], ~~~B_{0,1}[2]$ \\

2540: $(2,2,2,1)$    & 48    & $A_{0,1}[2]$  \\

2541: $(2,2,1,2)$    & 48     & $B_{1,0}[2]$  \\ \hline

2542: $(1,4,2,1)$    & 48     & $A_{0,1}[4]$  \\

2543: $(1,4,1,2)$    & 48     &  $B_{1,0}[4]$ \\

2544: \hline

2545: \end{tabular}

2546: \caption{Orbits of super-integral Descartes configurations classified by $\bg$.}

2547: ~\label{ta81}

2548: \end{center}

2549: \end{table} 

2550: 

2551: More details of this calculation are given in Table~\ref{ta81}.

2552: To explain the notation in Table~\ref{ta81}, 

2553: for any $4\times 4$ integral matrix $\bW$, let $g_i$ be the greatest common 

2554: divisor of entries $w_{1,i}$, $w_{2,i}$, $w_{3,i}$, $w_{4,i}$ 

2555: in the $i$-th column. 

2556: Then the action of $\sA^S$ preserves the vector $\bg=(g_1, g_2, g_3, g_4)$. 

2557: (For $\bW=\bW_\sD$, $g_2$ is the greatest common divisor of the curvatures.)  

2558: In each  row of the table  we give the number of orbits of $\sA^S$ formed by 

2559: the set of ordered, oriented  Descartes configurations that are super-integral

2560: with the given $\bg$. We also list the representatives of those orbits. 

2561: Note that each entry in the column labelled ``Representative'' 

2562: stands for 48 orbits, obtained 

2563: by taking two choices of (total) orientation, and 24 choices of permutation

2564: of rows. Also note the symmetry that the 

2565: configurations with $\bg$  are inverse of the ones with

2566: $\bg'=\bP_{12} \bg$.

2567: We  conclude from the table that the set of  ordered, oriented

2568: Descartes configurations that are super-integral comprise  

2569: $384(1+\frac{1}{2} +\frac{1}{4})=672$ orbits of the super-Apollonian group.

2570: ~~~$\bsq$

2571: 

2572: 

2573: 

2574: %****************************************************************************

2575: %

2576: % SECTION 9  Concluding remarks ?

2577: %

2578: %*****************************************************************************

2579: 

2580: \section{Concluding remarks}

2581: \setcounter{equation}{0}

2582: 

2583: This paper showed that the ensemble

2584: of all primitive,

2585: strongly integral Apollonian circle packings can  be

2586: simultaneously described 

2587: in terms of  an orbit of a larger discrete group,

2588: the super-Apollonian group, acting on the

2589: standard strongly-integral super-packing.  

2590: Study of the  locations of

2591: the individual integer packings inside 

2592: the standard  super-packing, presented in \S6,

2593:  leads  to interesting

2594: questions, not all of which are resolved.

2595: The standard super-packing also played a

2596: role in analyzing the structure of the

2597: super-Apollonian group as a discrete subgroup,

2598: carried out in  \S7.

2599: 

2600: The various illustrations show 

2601:  the usefulness of graphical representations, 

2602: as a guide to both finding and illustrating

2603: results. This contribution  is due in large part

2604: to the statistician co-authors (CLM and AW).

2605: Graphics were  particularly useful in finding extra

2606: symmetries of these objects, such as those illustrated

2607: in Figures 9--11 and subsequently proved by

2608: S. Northshield \cite{No03}.

2609: However one must not forget the

2610: adage of H. M. Stark \cite[p. 225]{St78}:

2611: ``A picture is worth a thousand words, provided 

2612: one uses another thousand words to justify

2613: the picture.''  Section 3 of this paper 

2614: provides such a justification for certain

2615: features of Figure 4.

2616: 

2617: There remain some open questions, particularly

2618: concerning the classification of all 

2619: integer root quadruples 

2620:  classified by fixed curvature $-N$ of the

2621: bounding circle. This quantity is known

2622: to be interpretable as a class number,

2623: as described in \cite[Theorem 4.2]{GLMWY2}.

2624: In \S6 we observed some symmetries of these

2625: root quadruples inside the standard super-packing,

2626: see Figures ~\ref{fig4}--\ref{fig6}.  

2627: There is a new invariant

2628: that can be associated to such quadruples,

2629: which is their nesting depth as defined in \S4

2630: with respect to the generating quadruple

2631: $\sD_0$ of the standard super-packing. It would

2632: be interesting to see whether this invariant

2633: might give some  further insight into class numbers. 

2634: 

2635:  

2636: 

2637: %****************************************************************************

2638: %

2639: % SECTION 10  Appendix: Strong Integrality Criterion 

2640: %

2641: %*****************************************************************************

2642: \newpage

2643: \section{Appendix.  Strong Integrality Criterion}

2644: % for  Descartes Configurations}

2645: \setcounter{equation}{0}

2646: 

2647: This appendix  establishes that to

2648: show a Descartes configuration is strongly integral

2649: it suffices to show that

2650:  three of its four circles are strongly integral.

2651: This affirmatively answers a question posed to us by K. Stephenson.

2652: 

2653: %

2654: %

2655: % Theorem 10.1

2656: %

2657: %

2658: 

2659: \begin{theorem}~\label{th101}

2660: A Descartes configuration $\sD$ has integral curvature-center

2661: coordinates $\bM_{\sD}$ if and only if it contains three

2662: circles having integer curvatures and whose

2663: curvature$\times$centers, viewed as complex numbers,  lie in $\zz[i]$.

2664: \end{theorem}

2665:                                                                                 

2666: \pf

2667: The condition is clearly necessary, and the problem is

2668: to show it is sufficient. We write the circle

2669: centers as complex numbers $\bz_j = x_j + i y_j$. So suppose $\sD$

2670: contains three circles with curvatures

2671: $b_1,~b_2, ~b_3 \in \zz$ and with

2672: curvature$\times$centers $b_1\bz_1,~b_2\bz_2,~b_3\bz_3 \in \zz[i]$.

2673: We must show that

2674: the fourth circle in the configuration has

2675: $b_4 \in \zz$ and $b_4\bz_4 \in \zz[i].$

2676: 

2677: For later use, we note that Theorem 3.1 of Part I \cite{GLMWY11} has a nice

2678: interpretation using complex numbers $\bz$

2679: to represent circle centers. This was formulated  in \cite{LMW02}

2680: as the Complex Descartes theorem. 

2681: It gives 

2682: \beql{901}

2683: b_1^2\bz_1^2 +  b_2^2\bz_2^2 + b_3^2\bz_3^2 +  b_4^2\bz_4^2 =

2684: \frac{1}{2}( b_1\bz_1 +  b_2\bz_2 + b_3\bz_3 +  b_4\bz_4)^2.

2685: \eeq

2686: and

2687: \beql{soddy}

2688: b_1^2\bz_1 +  b_2^2\bz_2 + b_3^2\bz_3 +  b_4^2\bz_4

2689: = \frac{1}{2} ( b_1\bz_1 +  b_2\bz_2 + b_3\bz_3 +  b_4\bz_4)

2690: (b_1 + b_2 +b_3 +b_4).

2691: \eeq

2692:                                                                                 

2693:   We claim  that $b_4 \in \zz.$ This is proved in

2694: the following two cases.

2695: 

2696: {\em Case 1. $b_1b_2b_3\neq 0$.}

2697:                                                                                 

2698: We first suppose

2699: that  $\bz_1=0$.

2700: If both $x_2$ and $x_3$ are zero, then $-\frac{1}{b_1} =\frac{1}{b_2}+\frac{1}{b_3}$, which means $b_1b_2+b_2b_3+b_3b_1=0$. Hence $b_4=b_1+b_2+b_3 \in \zz$.

2701: Otherwise by permuting $\bz_2$ and $\bz_3$ if

2702: necessary we may assume that $x_2 \ne 0$.

2703: Then the following equations encode the distance between 

2704: the circle centers, since the circles touch.

2705: \begin{eqnarray}

2706: x_2^2+y_2^2=(\frac{1}{b_1} + \frac{1}{b_2})^2, \label{xy1_sqr} \\

2707: x_3^2+y_3^2=(\frac{1}{b_1} + \frac{1}{b_3})^2, \label{xy_sqr} \\

2708: (x_3-x_2)^2+(y_3 - y_2)^2

2709: =(\frac{1}{b_2}+\frac{1}{b_3})^2. \label{xyxy1}

2710: \end{eqnarray}

2711: We wish to solve these equations for $y_3$ in terms of

2712: $b_1, b_2, b_3, x_1$ and $x_2$.

2713: To this end we subtract the

2714: first two equations from the third and obtain

2715: $$

2716: 2(x_2x_3 + y_2 y_3) = (\frac{1}{b_1} + \frac{1}{b_2})^2

2717: + (\frac{1}{b_1} + \frac{1}{b_3})^2 -(\frac{1}{b_2}+\frac{1}{b_3})^2 := R.

2718: $$

2719: Calling the right side of this equation $R$, we obtain

2720: $$

2721: x_3 = \frac{1}{2x_2}\left( R - 2y_2 y_3 \right).

2722: $$

2723: where the division  is allowed since $x_2 \ne 0$.

2724: Substituting this in the second equation yields a quadratic

2725: equation in $y_3$, with $x_3$ eliminated, namely

2726: (after multiplying by $4x_2^2$),

2727: $$

2728: 4(x_2^2 + y_2^2)  y_3^2 - 4R~ y_2y_3+R^2

2729: - 4x_2^2  (\frac{1}{b_1} + \frac{1}{b_3})^2 = 0.

2730: $$

2731: Since $y_3$ is

2732: rational, this  equation has rational  solutions, so the discriminant $\Delta$

2733: must be the square of a rational number. After some

2734: calculation one obtains

2735: \[

2736: \Delta=16^2 x_2^2 \left(\frac{ b_1b_2+b_1b_3+b_2b_3}{ (b_1b_2b_3)^2} \right).

2737: \]

2738: Since $x_2, b_1, b_2$ are

2739: nonzero it  follows that  $b_1b_2+b_2b_3+b_1b_3$ is a perfect square.

2740: Viewing  the Descartes equation as a quadratic equation in $b_4$

2741: we obtain the formula

2742: $b_4 :=b_1+b_2+b_3\pm2\sqrt{b_1b_2+b_2b_3+b_1b_3},$

2743: which shows that both roots are integers.

2744: These are the oriented curvatures of the

2745: two possible choices for the fourth circle in the

2746: Descartes configuration, so  $b_4 \in \ZZ$.

2747:                                                                                 

2748: Now assume that   $\bz_1= x_1 + i y_1$ is arbitrary. Define

2749: \begin{equation*}

2750: (s_2, t_2) := (x_2 - x_1, y_2 - y_1) ~~\mbox{and}~~ (s_3, t_3) := (x_3 - x_1,

2751: y_3-y_1).

2752: \end{equation*}

2753: Then $s_2, t_2, s_3, t_3$ are rational numbers, and they also satisfy

2754: equations \eqref{xy1_sqr}, \eqref{xy_sqr}, \eqref{xyxy1}. (just replace

2755: $x_2, y_2, x_3, y_3$ in \eqref{xy1_sqr}, \eqref{xy_sqr},

2756: \eqref{xyxy1} by $s_2,t_2,s_3, t_3$, respectively.)

2757: By the preceding argument, we again have

2758: $16s_2^2 (b_1b_2+b_2b_3+b_1b_3)/(b_1b_2b_3)^2 = q^2$

2759: for some rational number $q$.

2760: This implies that $b_1b_2+b_2b_3+b_1b_3$ is a perfect square and

2761: consequently that

2762: $b_4$ is an integer.

2763:                                                                                 

2764: {\em Case 2. $b_1b_2b_3=0$.}

2765: This  is proved similarly,  with an

2766: easier calculation.

2767:                                                                                 

2768: The claim follows, so $b_4 \in \ZZ$.

2769: 

2770: We now proceed to  show that

2771: $b_4\bz_4 \in \zz[i]$. Now  \eqn{901}  gives

2772: \begin{equation}~\label{reduce0}

2773: b_4\bz_4=b_1\bz_1+b_2\bz_2+

2774: b_3\bz_3\pm2\sqrt{b_1\bz_1b_2\bz_2+b_2\bz_2b_3\bz_3+b_1\bz_1b_3\bz_3}.

2775: \end{equation}

2776: The equation  \eqn{soddy} gives

2777: \begin{equation}~\label{reduce}

2778: (b_1+b_2+b_3-b_4)b_4\bz_4=2(b_1^2\bz_1+b_2^2\bz_2+b_3^2\bz_3)-(b_1+b_2+b_3+b_4)

2779: (b_1\bz_1+b_2\bz_2+b_3\bz_3).

2780: \end{equation}

2781: We treat two mutually exhaustive cases.

2782:                                                                                 

2783: {\em Case 1.  $b_1+b_2+b_3 \neq b_4.$}

2784:                                                                                 

2785: Then \eqn{reduce} gives

2786: $b_4\bz_4=x_4+iy_4$ for some rational

2787: numbers $x_4, y_4$. However

2788:   $b_1\bz_1, b_2\bz_2, b_3\bz_3$ are integers by hypothesis,

2789: whence \eqn{reduce0} shows that $b_4z_4$ is an algebraic integer.

2790: Since $x_4, y_4$  are rational, we conclude that

2791: $b_4\bz_4$ must be an integer, i.e., in $\zz[i]$.

2792:                                                                                 

2793: {\em Case 2.  $b_1+b_2+b_3=b_4$.}

2794:                                                                                 In this case, we have

2795: $b_1b_2+b_2b_3+b_1b_3 =0$. Now from

2796: \eqref{reduce}, we have

2797: \[

2798:  b_1^2\bz_1+b_2^2\bz_2+b_3^2\bz_3=(b_1+b_2+b_3)(b_1\bz_1+b_2\bz_2+b_3\bz_3),

2799: \]

2800: which can be simplified to

2801: \[

2802: b_2b_3\bz_1+b_1b_3\bz_2+b_1b_2\bz_3=0.

2803: \]

2804: Thus

2805: \begin{eqnarray*}

2806: (b_1\bz_1)(b_2\bz_2)+(b_2\bz_2)(b_3\bz_3)+(b_3\bz_3)(b_1\bz_1) & = &

2807: b_1(b_2\bz_2+b_3\bz_3)\frac{b_1b_3\bz_2+b_1b_2\bz_3}{-b_2b_3}

2808: +(b_2\bz_2)(b_3\bz_3) \\

2809: &=&-b_1^2\bz_2^2-b_1^2\bz_3^2+\bz_2\bz_3 \left(b_2b_3-

2810: \frac{b_1^2}{b_2b_3}(b_2^2+b_3^2)\right) \\

2811: &=& -b_1^2(\bz_2-\bz_3)^2.

2812: \end{eqnarray*}

2813: Now $b_1(\bz_2-\bz_3)$ is a Gaussian

2814: rational number whose square is integral.

2815: Hence $b_1(\bz_2-\bz_3)$ must be integral.

2816: It follows that $b_4\bz_4=b_1\bz_1+b_2\bz_2+b_3\bz_3\pm 2b_1(\bz_2-\bz_3)i$ is

2817: in $\zz[i]$.~~~$\bsq$ \\

2818:                                                                                 

2819: Since the Apollonian group consists of integer matrices,

2820: all Descartes configurations

2821: in an  Apollonian packing generated by  a strongly integral

2822: Descartes configuration are  strongly

2823: integral.  This explains

2824: the integrality properties of curvatures and

2825: curvature $\times$ center pictured in the packing

2826: in \S1, for example. The previous result now  gives

2827: a weaker necessary and sufficient condition for an Apollonian

2828: packing to be strongly integral.

2829: 

2830: %

2831: % Theorem 10.2

2832: %

2833: 

2834: \begin{theorem} \label{th32} 

2835: An Apollonian circle packing is strongly integral if

2836: and only if it contains three mutually tangent

2837: circles which have integer curvature-center

2838: coordinates.

2839: \end{theorem}

2840:                                                                                 

2841: \pf

2842: Suppose we are give three mutually tangent circles

2843: in the packing that are strongly integral.

2844: Any set of three mutually tangent circles in

2845: the packing is part of some Descartes configuration

2846: in this packing. This follows from the recursive

2847: construction of the packing, which has a finite

2848: number of circles at each iteration. If iteration $j$

2849: is the first iteration at which all three tangent

2850: circles are present, at that iteration they

2851: necessarily belong to a unique Descartes configuration.

2852: Theorem~\ref{th101} now implies that this

2853: Descartes configuration is strongly integral.

2854: It then follows that the whole Apollonian

2855: packing is strongly integral. This proves the

2856: ``if'' direction, and the ``only if'' direction

2857: is immediate.

2858: ~~~$\bsq$

2859:                                                                                 

2860: 

2861: 

2862: %*****************************************************************************

2863: %

2864: %  References 

2865: %

2866: %*****************************************************************************

2867: % 

2868: 

2869: 

2870: \clearpage

2871: \begin{thebibliography}{99}

2872: 

2873: \bibitem{AS97}

2874: D. Aharonov and K. Stephenson,

2875: Geometric sequences of discs in the

2876: Apollonian packing.

2877: Algebra i Analiz {\bf 9} (1997), No. 3, 104--140.

2878: [English version: St. Petersburg Math. J. {\bf 9} (1998), 509--545.]

2879: 

2880: \bibitem{Be83}

2881: A. F. Beardon,

2882: {\em The Geometry of Discrete Groups,}

2883: Springer-Verlag: New York 1983.

2884: 

2885: \bibitem{GLMWY11}

2886: R. L. Graham, J. C. Lagarias, C. L. Mallows, A. Wilks and C. Yan,

2887: Apollonian circle packings: geometry and group theory

2888: I. The Apollonian group, preprint.

2889: eprint: {\tt arXiv math.MG/0010298}.

2890: 

2891: \bibitem{GLMWY13}

2892: R. L. Graham, J. C. Lagarias, C. L. Mallows, A. Wilks and C. Yan,

2893: Apollonian circle packings: geometry and group theory

2894: III. Higher dimensions, 

2895: eprint: {\tt arXiv math.MG/0010324}.

2896: 

2897: \bibitem{GLMWY2}

2898: R. L. Graham, J. C. Lagarias, C. L. Mallows, A. Wilks and C. Yan,

2899: Apollonian circle packings: number theory, 

2900: Journal of Number Theory, {\bf 100} no.1 (2003), 1--45. 

2901: [eprint: arXiv math.NT/0009113]

2902: 

2903: \bibitem{LMW02}

2904: J. C. Lagarias, C. L. Mallows and A. Wilks,

2905: Beyond the  Descartes circle theorem,

2906: Amer. Math. Monthly {\bf 109} (2002),  338--361.

2907: [eprint: {\tt arXiv math.MG/0101066}]

2908: 

2909: \bibitem{M82}

2910: B. B. Mandelbrot,

2911: {\em The Fractal Geometry of Nature},

2912: Freeman: New York, 1982.

2913: 

2914: 

2915: 

2916: \bibitem{Max82}

2917: G. Maxwell,

2918: Sphere packings and hyperbolic reflection groups.

2919: J. Algebra {\bf 79} (1982), 78--97.

2920: 

2921: 

2922: \bibitem{MSW02}

2923: D. Mumford, C. Series, and D. Wright,

2924: {\em Indra's Pearls: The Vision of Felix Klein},

2925: Cambridge U. Press: Cambridge 2002.

2926: 

2927: \bibitem{No03}

2928: S. Northshield,

2929: On Apollonian circle packings,

2930: preprint.

2931: 

2932: 

2933: %\bibitem{Ru70}

2934: %W. R\"uhl,

2935: %{\em The Lorentz group and harmonic analysis},

2936: %W. A. Benjamin: New York 1970.

2937: 

2938: \bibitem{So92}

2939: B. S\"oderberg,

2940: Apollonian tiling, the Lorentz group, and regular trees.

2941: Phys. Rev. A {\bf 46} (1992), No. 4, 1859--1866.

2942: 

2943: \bibitem{St78}

2944: H. M. Stark,

2945: {\em An Introduction to Number Theory}, 

2946: M. I. T. Press: Cambridge-London 1978. 

2947: 

2948: \bibitem{Wi81}

2949: J. B. Wilker,

2950: Inversive Geometry,

2951: in: {\em The Geometric Vein}, (C. Davis, B. Gr\"unbaum, F. A. Sherk, Eds.),

2952: Springer-Verlag: New York 1981, pp. 379--442.

2953: 

2954: \end{thebibliography}

2955: 

2956: {\tt

2957: \begin{tabular}{lllll}

2958: email: & graham@ucsd.edu \\

2959: & lagarias@umich.edu \\

2960:   & colinm@research.avayalabs.com \\

2961:  & allan@research.att.com \\

2962: & cyan@math.tamu.edu

2963: \end{tabular}

2964:  }

2965: 

2966: 

2967: 

2968: \end{document}

2969: 

2970: 

2971: 

2972: 

2973: 

2974: 

2975: 

2976: 

2977: 

2978: 

2979: 

2980: 

2981: 

2982: 

2983: 

2984: 

2985: