1: \documentclass[12pt]{amsart}

2: \input epsf.tex

3: \usepackage{latexsym}

4: \usepackage{amsmath,amsthm,amsfonts,amssymb}

5: \usepackage{epsfig}

6: 

7: \setlength{\topmargin}{0cm}

8: \setlength{\headsep}{1cm}

9: \setlength{\evensidemargin}{.0cm}

10: \setlength{\oddsidemargin}{.0cm}

11: \setlength{\textheight}{21.cm}

12: \setlength{\textwidth}{16.6cm}

13: 

14: \newtheorem{pro}{Proposition}[section]

15:  \newtheorem{thm}[pro]{Theorem}

16:  \newtheorem{lem}[pro]{Lemma}

17:  \newtheorem{defn}[pro]{Definition}

18:  \newtheorem{clm}[pro]{Claim}

19:  \newtheorem{example}[pro]{Example}

20:  \newtheorem{cnj}[pro]{Conjecture}

21:  \newtheorem{cor}[pro]{Corollary}

22:  \newtheorem{question}[pro]{Question}

23:  \newtheorem{remark}[pro]{Remark}

24: 

25: %misc

26: \def\A{{\cal A}}

27:  \def\B{{\cal B}}

28:  \def\C{{\mathbb C}}

29: \def\c{{\cal C}}

30: \def\E{{\cal E}}

31:  \def\H{{\mathbb H}}

32:  \def\G{{\cal G}}

33: \def\g{{\it {\bf g}}}

34:  \def\P{{\cal P}}

35:  \def\O{{\rm O}}

36: \def\oo{{\widetilde O}}

37: \def\Per{{\rm Per}}

38: 

39:  \def\M{{\cal M}}

40: \def\m{{\bar M}}

41:  \def\F{{\mathbb F}}

42:  \def\T{{\cal T}}

43: \def\tt{{\widetilde T}}

44:  \def\N{{\mathbb N}}

45:  \def\R{{\mathbb R}}

46:  \def\chix{{\raise.5ex\hbox{$\chi$}}}

47:  \def\d{{\cal d}}

48: \def\Z{{\mathbb Z}}

49: 

50: %mathbb

51: \def\mE{{\mathbb E}}

52: 

53: 

54: %bar

55: \def\bP{{\bar P}}

56: \def\bgamma{{\bar \gamma}}

57: 

58: %script

59: \def\sO{{\cal O}}

60: \def\sF{{\cal F}}

61: \def\sP{{\cal P}}

62: \def\sG{{\cal G}}

63: \def\sT{{\cal T}}

64: \def\sN{{\cal N}}

65: \def\sG{{\cal G}}

66: \def\sH{{\cal H}}

67: \def\sS{{\cal S}}

68: \def\sA{{\cal A}}

69: \def\sB{{\cal B}}

70: \def\sC{{\cal C}}

71: \def\sX{{\cal X}}

72: \def\sY{{\cal Y}}

73: \def\sK{{\cal K}}

74: \def\sW{{\cal W}}

75: \def\sV{{\cal V}}

76: \def\sU{{\cal U}}

77: \def\sI{{\cal I}}

78: \def\sM{{\cal M}}

79: %manifolds

80: 

81: \def\mM{{\bf M}}

82: \def\mO{{\bf O}}

83: \def\mP{{\bf P}}

84: \def\mT{{\bf T}}

85: 

86: %isometries

87: 

88: 

89: %geodesics in \H^3

90: \def\tA{{\widetilde A}}

91: \def\tB{{\widetilde B}}

92: \def\tN{{\widetilde N}}

93: \def\tT{{\widetilde T}}

94: \def\tF{{\widetilde F}}

95: \def\tH{{\widetilde H}}

96: \def\tG{{\widetilde G}}

97: \def\tP{{\widetilde P}}

98: \def\tO{{\widetilde O}}

99: \def\tM{{\widetilde M}}

100: \def\tC{{\widetilde C}}

101: \def\tX{{\widetilde X}}

102: \def\tY{{\widetilde Y}}

103: \def\tK{{\widetilde K}}

104: \def\tW{{\widetilde W}}

105: \def\tS{{\widetilde S}}

106: \def\tgamma{{\widetilde \gamma}}

107: \def\tsigma{{\widetilde \sigma}}

108: \def\tp{{\widetilde p}}

109: \def\tq{{\widetilde q}}

110: \def\tv{{\widetilde v}}

111: \def\tw{{\widetilde w}}

112: %geodesics on a closed surface

113: 

114: \def\pG{{\bar G}}

115: \def\pH{{\bar H}}

116: \def\pB{{\bar B}}

117: \def\pA{{\bar A}}

118: \def\pT{{\bar T}}

119: \def\pP{{\bar P}}

120: \def\pO{{\bar O}}

121: \def\pS{{\bar S}}

122: 

123: %hats

124: \def\hH{{\hat H}}

125: \def\hK{{\hat K}}

126: 

127: 

128: %widths

129: \def\wT{{T}}

130: \def\wF{{F}}

131: \def\wH{{H}}

132: \def\wG{{G}}

133: \def\wP{{P}}

134: \def\wO{{O}}

135: \def\wN{{N}}

136: \def\wA{{A}}

137: \def\wB{{B}}

138: \def\wC{{C}}

139: \def\wX{{X}}

140: \def\wY{{Y}}

141: \def\wK{{K}}

142: 

143: %\misc

144: \def\Tree{{Tree}}

145: \def\Free{{F}}

146: \def\Rules{{Rules}}

147: \def\Tilings{{Tilings}}

148: \def\Frames{{Frames}}

149: \def\bF{{\mathbb F}}

150: \def\f{{f}}

151: \def\Sp{{Sp}}

152: \def\sech{{\textnormal{sech}}}

153: \def\csch{{\textnormal{csch}}}

154: \def\arccosh{{\textnormal{arccosh}}}

155: \def\arcsinh{{\textnormal{arcsinh}}}

156: \def\gM{{M}}

157: \def\m{{\hat M}}

158: \def\ep{{\varepsilon}}

159: 

160: \def\Sym(#1){\Sigma_{#1}}

161: \def\limmu_#1{\mu^{(#1)}}

162: \def\cvxmu_#1{\lambda_{#1}}

163: \def\infrac#1#2{(#1)/(#2)}

164: \def\Seq#1{\langle #1 \rangle}

165: \def\uP{\widehat P}

166: \def\II#1{{\bf 1}_{\{#1\}}}

167: \def\msnote#1{{\bf ** #1 **}}

168: 

169: 

170: \begin{document}

171: \title{A Solidification Phenomenon in Random Packings}

172: \author{L. Bowen, R. Lyons, C. Radin, \and P. Winkler}

173: \address{Department of Mathematics\\ Indiana University\\ Bloomington,

174:   IN 47405}

175: \email{lpbowen@indiana.edu}

176: \address{Department of Mathematics\\ Indiana University\\ Bloomington,

177:   IN 47405}

178: \email{rdlyons@indiana.edu}

179: \address{Department of Mathematics\\ University of Texas\\ Austin, TX 78712}

180: \email{radin@math.utexas.edu}

181: \address{Department of Mathematics\\ Dartmouth College\\ Hanover, NH 03755}

182: \email{peter.winkler@dartmouth.edu}

183: \thanks{Research was partially supported by the National Science

184: Foundation under grants DMS-0406017 and DMS-0352999. The authors also 

185: thank the Banff International Research Station for support at a workshop

186: where this research was begun.}

187: \date{}

188: 

189: \begin{abstract}

190: We prove that uniformly random packings of copies of a certain simply-connected

191: figure in the plane exhibit

192: global connectedness at all sufficiently high densities,

193: but not at low densities.

194: \end{abstract}

195: \maketitle

196: %\section{Notes}

197: 

198: % Should we say ``infinitely connected component of loose-linked

199: % tiles'' or ``infinite linked component'' or ``infinite linked cluster''?

200: 

201: 

202: 

203: \section{Introduction}

204: 

205: The densest way to cover a large area with non-overlapping unit disks

206: is as in Figure~\ref{fig:disks}.

207: 

208: \begin{figure}[htp]

209: \epsfxsize=2.8truein

210: $$\epsfbox{disks2.eps}$$

211: \caption{The densest packing of unit disks in the plane}\label{fig:disks}

212: \end{figure}

213: 

214: \noindent

215: A {\bf packing} is a

216: collection of congruent copies of a subset with pairwise disjoint interiors.

217: See \cite{Fej}

218: for a proof that the above packing is indeed densest possible for unit disks.

219: 

220: It is an old unsolved problem to understand whether densest packings

221: of spheres, simplices or other shapes,

222: in a Euclidean or hyperbolic space of any dimension,

223: exhibit

224: crystallographic symmetry.  For instance, this is the spirit of Hilbert's

225: eighteenth problem; see \cite{Fej, Rad} for background. 

226: 

227: Using physics models of two- and three-dimensional matter as a guide, we are

228: tempted to try to gain insight about densest packings by considering

229: packings at 

230: densities below the maximum. (For an example concerning

231: spheres in $\R^3$, see \cite{KRS}.) In effect, we are emphasizing

232: not so much that densest packings and sparse packings differ by their

233: {\em symmetry}, as that they differ in some fundamental geometric

234: fashion. Indeed, it is

235: commonly suggested in the physics literature (see for instance

236: \cite{AH}) that 2-dimensional models of 

237: matter do not exhibit crystallographic symmetry, and it is sometimes

238: said by mathematicians that in high dimensional Euclidean space,

239: densest packings of spheres may not have crystallographic symmetry. So

240: perhaps it is appropriate to reexamine the precise manner in which

241: densest packings differ fundamentally from sparse packings, and to use

242: packings at less than optimum density as a guide.

243: 

244: The density of the unit disk packing of Figure~\ref{fig:disks} is $\widetilde d:=

245: {\pi/ \sqrt{12}}\approx 0.91$; packings of density

246: 0.89 (say) can be obtained by shrinking the disks slightly, shaking

247: the whole collection, and then

248: expanding the whole picture to recover unit-size disks.  

249: One might ask whether by doing this one preserves long-range correlations.

250: Experiments and discrete models indicate that such long-range

251: correlations

252: are indeed preserved, but this has never been proved.

253: See \cite{BLRW} for background.

254: 

255: In this work we prove the conjectured behavior, not for disks but for deformed

256: disks, copies of a ``zipper'' tile designed specifically for the

257: purpose; see Figure \ref{fig:tile}.

258: This tile can cover the plane completely, in which case the packing has

259: density 1, and is completely connected in any sense.

260: What we show is that even at somewhat lower densities, the uniform random

261: packing still has rich structure; in particular it has a form of

262: connectedness associated with site percolation \cite{G}.

263: What this means for packing large but finite boxes (with torus boundary

264: conditions) is that the necessary gross irregularities of most packings at such

265: high densities occur, but not in a way to disconnect the packings.

266: Although we define ``uniform random packing" of the plane by limits of

267: measures on packings of finite boxes, the key to our proof is to examine

268: isometry-invariant probability measures on packings of the whole plane and

269: to show that the ones that maximize ``degrees of freedom per tile"

270: are unique for high densities.

271: 

272: In physics one is interested in change of behavior as density

273: changes.

274: We show that at high density there is a nonzero probability of

275: an infinite linked component, and that this probability is zero at

276: low density.

277: Thus, there are

278: different ``phases'' of the packings \cite{BLRW}.

279: (This is the same phase transition seen in continuum percolation, where one

280: looks at overlapping disks with random independent centers, but our methods

281: are quite different. Indeed, no such result is known for packings of disks.)

282: 

283: Although we believe such a result also holds for packings of

284: disks or of spheres --- pairs of which would be called ``linked''

285: if sufficiently close --- we are able to prove the result only for our tiles,

286: which are shaped to allow three well-defined levels of pairwise

287: separation. It is generally understood that crystalline behavior

288: is not seen in two dimensions, so the form of connectedness we use

289: may be useful in understanding the role of geometry in Hilbert's problem.

290: 

291: 

292: \section{Description of the tile}

293: 

294: We consider packings by a deformed disk denoted by $t$,

295: referred to as ``the tile'' and depicted in Figure \ref{fig:tile}. In this

296: section, we define it precisely.

297: 

298: Let $H$ be a regular hexagon of area 1. Let $r$ be the radius of the in-circle of $H$.

299: Let $D$ be a disk concentric with the in-circle and of radius $r+\rho$,

300: where $0<\rho\ll 1$

301: is a number we shall choose more precisely later. We shall construct the shape $t$  by modifying $H$ as follows; $D$ will be called

302: the {\bf shadow disk} of $t$.

303: 

304: \begin{figure}[htp]

305: \epsfxsize=2.7truein

306: $$\epsfbox{nex3.ps}$$

307: %\centerline{\epsfig{figure=hex10af.ps}}

308: %\vskip-.5truein

309: \caption{The zipper tile}\label{fig:tile}

310: \end{figure}

311: 

312: \begin{figure}[htp]

313: \epsfxsize=3.7truein

314: $$\epsfbox{hex3f.ps}$$

315: %\centerline{\epsfig{figure}}

316: %\vskip1.8truein

317: \caption{Close-up view of the fringe}\label{fig:measure}

318: \end{figure}

319: 

320: As shown in Figure \ref{fig:tile}, the tile $t$ equals $H$ with each side modified

321: by a ``fringe'' and each corner modified by a hook and inlet, where a

322: hook is about

323: half an element of the

324: fringe. As shown in Figure \ref{fig:measure}, the fringe height

325: is $2\rho$.  The elements of the fringe have two different size ``necks'', one of size

326: $\rho^2$ and one of size $2\rho^2$, allowing neighboring tiles to be linked in either

327: of two well-defined modes, ``tight linked'' and ``loose linked'', the

328: former illustrated

329: in Figure \ref{fig:tight} and  the latter illustrated

330: in Figure \ref{fig:loose}.  

331: We say that two tiles $t$ are

332: {\bf linked} (tightly or loosely) if when one is held fixed, the

333: other can be moved continuously only by a bounded amount (without

334: overlapping the first).  

335: A {\bf tight} link is one that permits no movement of one tile while fixing

336: the other, while a link that is not tight is called {\bf loose}.

337: A key feature of our model is that when two tiles are tightly

338: linked, any motion of one would necessitate a corresponding motion of the

339: other.  As we shall explain, the uniform

340: probability distribution on packings of the plane at given density is a

341: limit of such a distribution on packings of larger and larger tori.  In our

342: model,

343: these distributions on packings of finite tori are concentrated on packings with

344: the maximal number of degrees of freedom, and therefore intuitively the

345: fewest possible number of 

346: tiles bound by tight links.  This gives us useful control on the packings in the support

347: of our distributions.

348: 

349: \begin{figure}[htp]

350: \epsfxsize=3.5truein

351: $$\epsfbox{nex5.ps}$$

352: %\centerline{\epsfig{figure=hex7.ps}}

353: \caption{Tight-linked tiles}\label{fig:tight}

354: \end{figure}

355: 

356: \begin{figure}[htp]

357: \epsfxsize=3.5truein

358: $$\epsfbox{nex6.ps}$$

359: %\centerline{\epsfig{figure=hex6.ps}}

360: \caption{Loose-linked tiles}\label{fig:loose}

361: \end{figure}

362: 

363: A tile is called {\bf fully linked on one side} if it is linked with

364: another tile on that side in such a way that either

365: they are tight linked and the line joining their centers goes through the

366: midpoint of the sides of the corresponding hexagons; or

367: the tiles are not tight linked but can be moved continuously so that their

368: shadow disks touch each other.

369: A tile is {\bf fully linked} if it is fully linked on all sides.

370: We note that the fully tight-linked packing (Figure~\ref{fig:tight}) corresponds to a

371: tiling by the original hexagon and has density 1, and that the tile has area 1 by construction.

372: 

373: 

374: 

375: \section{Statement of Results}

376: 

377: To state our results we need some notation. Let $X$ be the space of

378: all packings of the plane by the tile $t$. Given a compact subset $K$ of the

379: plane and two packings of the plane, we consider the distance between the

380: two packings with respect to $K$ to be the Hausdorff distance between the

381: unions of the tiles in the respective packings intersected with $K$.

382: Then $X$ is endowed with the

383: topology of Hausdorff convergence on compact subsets; $X$ is compact.

384: Intuitively, two

385: packings are close in $X$ if they are close in the Hausdorff sense in

386: a large ball centered at the origin. We shall define a probability

387: measure on $X$ that is ``uniform'' on the set of all packings of a

388: fixed density. For this, we shall need the space $X_n$ of all packings

389: by the tile of the $n \times n$ torus $\R^2/(n\Z)^2$.

390: 

391: For any integer $m$, let $X_{n,m} \subset X_n$ consist of those packings which

392: contain exactly $m$ tiles ($X_{n,m}$ is empty if $m$ is large

393: enough). To each tile, we assign the set of $6$ unit vectors based on

394: its center and pointing to the center of each of its edges. Through

395: this assignment, we can view $X_{n,m}$ as a subset of $T_n^m/\Sym(m)$,

396: where $T_n$ is the unit tangent bundle of the $n\times n$ torus modulo

397: a $2\pi/6$ rotation and the symmetric group $\Sym(m)$ acts by permuting

398: the factors.

399: 

400: When $m/n^2$ is small, $X_{n,m}$ is $(3m)$-dimensional. However, when $m/n^2$ is

401: sufficiently large, the dimension of $X_{n,m}$ inside $T_n^m/\Sym(m)$

402: is less than $3m$. This is because at least two tiles in any

403: packing of $X_{n,m}$ will have to be tightly linked, so that it is

404: impossible to move one continuously without moving the other. Thus it is

405: useful to decompose

406: $X_{n,m}$ into a (finite) disjoint union of sets $X_{n,m,k}$ of

407: packings containing exactly $k$ tight links. Generically, the

408: dimension of $X_{n,m,k}$ is $3(m-k)$. The dimension can be strictly

409: less than this if the packings are jammed in the sense of \cite{DTSC},

410: although this fact will not be important for us. The {\bf top dimension} of

411: $X_{n, m}$ means the maximum dimension of all $X_{n,m,k}$. Let

412: $\mu_{n,m}$ be the probability measure on $X_{n,m}$ obtained by

413: normalizing the Hausdorff measure on $X_{n,m}$ in the top dimension of

414: $X_{n,m}$ with respect to the natural

415: metric inherited from $T_n^m/\Sym(m)$. We interpret $\mu_{n,m}$ as

416: being a uniform measure. The fact that $\mu_{n,m}$ is supported on

417: those packings with the fewest number of tight links will be crucial in

418: the analysis to follow.

419: 

420: Let $\tX_{n}$ be the space of all $(n\times n)$-periodic packings of the

421: plane. In other words, $\tX_{n}$ consists of those packings that are

422: preserved under

423: translations by $n\Z \times n \Z$. Under the quotient map, this space

424: is naturally identified with $X_{n}$. Therefore, we can view the

425: measures $\mu_{n,m}$ as living on $\tX_n \subset X$.

426: 

427: For a fixed density $d \in [0,1]$, let $\limmu_d$ be any measure obtained

428: as the weak* limit of measures of the form $\mu_{n,m}$ such that $n

429: \to \infty$ and $m/n^2 \to d$. (Note that $m/n^2$ is the density of every

430: packing in the support of $\mu_{n,m}$ and $d$ is the average density

431: of a packing chosen with respect to $\limmu_d$: see Lemma \ref{lem:cont}.)

432: A priori, $\limmu_d$ may not

433: be unique, although we shall prove that it is for large enough $d$.

434: %Throughout this article when we speak of the probability of an event

435: %at density $d$, we mean with respect to $\limmu_d$.

436: 

437: As we have seen, two tiles $t$ can be

438: linked.  Therefore it makes sense to speak of a {\bf linked

439: component} of a packing; it is a maximal sub-packing such that for every two

440: tiles $t,t'$ in it, there is a sequence $t=t_1, t_2,\ldots,t_n=t'$ such

441: that $t_i$ is linked to $t_{i+1}$ ($i=1,\ldots,n-1$). A {\bf tight-linked component} is defined similarly, but we require $t_i$ to be tightly-linked to $t_{i+1}$.

442: 

443: 

444: We say that a measure on the space $X$ of packings is invariant if it preserved under the full isometry group of the plane. All the measures we consider are probability measures unless stated otherwise. 

445: 

446: Let $\lambda_0$ be the unique invariant measure on tilings (packings that

447: cover $\R^2$)

448: by our tile.  

449: Let $\lambda_1$ be the unique invariant measure

450: on packings by $t$ such that all tiles are fully loose linked, are as close

451: as possible to each other, and the packing has hexagonal symmetry. 

452: Write $\cvxmu_s:=s \lambda_1 + (1-s)\lambda_0$.

453: 

454:  Our main results are the following:

455: 

456: 

457: \begin{thm}\label{thm:main1} There exists $0<d_1 < 1$ such that if $d

458:   \ge d_1$, $\limmu_d$ is unique and equals $\cvxmu_s$,

459:   where $s:=\infrac{1-d}{ 1-d_1}$.

460: \end{thm}

461: 

462: \begin{cor}

463:   The $\limmu_d$-probability that the origin is inside a tile belonging to

464:   an infinite linked component is nonzero for $d \ge d_1$.

465: \end{cor}

466: 

467: \begin{pro}\label{thm:low1} For some $d_2>0$, the probability (with

468:   respect to any $\limmu_d$ for any $d < d_2$) that the origin

469: is inside an infinite linked component is zero.

470: \end{pro}

471: 

472: 

473: \section{Tile properties}

474: 

475: \begin{lem}

476:   For small $\rho$, 

477:   if tiles $t_1$ and $t_2$ are not tightly linked and do not overlap, then the

478:   distance between their centers is at least $2r + 2\rho$.

479: \end{lem}

480: \begin{proof}

481:   Consider the line segment from the center of $t_1$ to the center of

482:   $t_2$. If this segment traverses near a corner of $t_1$ or $t_2$,

483:   then it must be longer than $2r+2\rho$ for small enough $\rho$.

484:   Supposes it crosses a fringe of $t_1$ and of $t_2$. If the tiles

485:   are not linked, then the claim is obvious. If they are linked, then to

486:   minimize the distance, it must be that their fringes match up (so

487:   they are fully linked on one side). Thus the closest they can come

488:   is if the two are pushed flat up against each other so that their shadow

489:   disks touch. In this case, the distance between the centers is exactly

490:   $2r+\rho$.

491: \end{proof}

492: 

493: We shall say that two tiles are {\bf densely loose linked} if they are

494: loose linked and their shadow disks touch.

495: There is a unique invariant measure on maximally

496: dense packings by congruent disks \cite{BHRS}.

497: Hence the probability measure $\lambda_1$ that we defined earlier is the

498: unique invariant measure on packings by $t$ such that all tiles are fully

499: and densely loose linked. Let $d_1$ be the density of such a packing.

500: 

501: Given a tile $t$ in a packing $P$, we denote by $V(t)$ the Voronoi cell of

502: the center of $t$ with respect to the centers of the other tiles; that

503: is, $V(t)$ is the open set of points closer to the center of $t$ than to the

504: center of any other tile.

505:   We denote the area of a region $A$ of the plane by $|A|$.

506: 

507: \begin{lem}\label{lem:lambda1}

508:   The following holds for small enough $\rho>0$. For any packing $P$,

509:   if $t \in P$ is a tile that has no tight links, then the area of

510:   $V(t)$ is least $1/d_1$. Moreover, equality holds iff the configuration of

511:   tiles determining $V(t)$ is congruent to a corresponding

512:   configuration of a packing in the support of $\lambda_1$.

513: \end{lem}

514: 

515: \begin{proof}

516:   For a tile $t$, let $H(t)$ denote the hexagon from which $t$ is created.

517:   For $x > 0$, let $H_x(t)$ denote the homothetic copy

518:   $\frac{r+x}{r}H(t)$ about the center of $H(t)$.

519: 

520:   Suppose $t$ is a tile of $P$ without any tight links.  Consider

521:   the rays $R_1,\dots,R_6$ from the center of the hexagon $H(t)$ through

522:   each of its 6 vertices. These rays divide the plane into 6 sectors,

523:   $S_1,\dots,S_6$. 

524:   

525:    By construction, if $t$ and $t_1$ are

526:   loose linked, then $|H_{\rho}(t) \cap H_\rho(t_1)| = \O(\rho^2)$:

527:   The hexagon interiors do not intersect if they are parallel, while

528:   if they are not parallel, they can intersect only very slightly at

529:   a corner. The openings at

530:   which a corner can enter have area $\O(\rho^2)$ as $\rho\to 0$.

531: 

532:   Thus, we have proved that whenever $t$ is

533:   loose linked in the sector $S_i$, then $|V(t) \cap S_i| \ge

534:   |H_\rho(t)|/6 - \delta_1$, with $\delta_1 =\O(\rho^2)$ as $\rho\to 0$.

535:   

536:   Similarly, if $t$ and $t_2$ are not linked at all, then $|H_{2\rho}(t)

537:   \cap H_{2\rho}(t_2)| = \O(\rho^2)$:  Again, their

538:   interiors do not intersect if they are parallel, while if they

539:   are not parallel, they can intersect only very slightly at a

540:   corner. So 

541:   there exists $\delta_2>0$ such that whenever $t$

542:   is not linked in the sector $S_i$, we have $|V(t)\cap S_i| \ge |H_{2\rho}(t)|/6 -

543:   \delta_2$, with $\delta_2 =\O(\rho^2)$ as $\rho\to 0$.  

544:   

545:   Therefore, if $t$

546:   has no tight links but is not fully linked, then

547: $$

548: |V(t)| \ge j\Big( \frac{|H_\rho(t)|}{6} -\delta_1\Big) +

549: (6-j)\Big(\frac{|H_{2\rho}(t)|}{6}-\delta_2\Big)

550: $$

551: for some $j$ with $0 \le j \le 5$. Given that $\delta_1,\delta_2$ are

552: of order $\rho^2$ while $|H_{2\rho}(t)|-|H_{\rho}(t)|$ is of order

553: $\rho$, for $\rho$ small enough we may conclude that $|V(t)| >

554: |H_\rho(t)|$ in this case. 

555: 

556: On the other hand, the geometry of a tile is such that for small $\rho$, if

557: $t_1$ and $t_2$ are two tiles loose linked to $t$, then $t_1$ cannot be

558: tight linked to $t_2$. Now suppose that $t$ is fully loose linked. Then

559: the Voronoi cell of the center of $t$ is determined by six tiles

560: $t_1,\dots,t_6$ all loose linked to $t$ and all with the property that

561: their shadow disks $D,D_1,\ldots,D_6$ do not overlap (by the previous lemma). It follows 

562: \cite{Fej2} that $|V(t)| \ge |H_\rho(t)|$, with equality iff each of the disks

563: $D_1,\dots,D_6$ touches $D$. But there is only one way in which this can

564: occur (up to isometry). So $V(t)=H_\rho(t)$ in this case. This implies that the configuration

565: $t,t_1,\dots,t_6$ is congruent to a corresponding configuration of a

566: packing in the support of $\lambda_1$.

567: \end{proof}

568: 

569: It is easy to see that given $\rho > 0$, there exists $\ep>0$ such that for any finite

570: component $c$ of tight-linked tiles in any packing, the union $V_c$ of the Voronoi cells

571: of the centers of the tiles of $c$ has area at least $j_c +\ep \Per_c$.

572: Here $j_c$ is the number of tiles in $c$ and $\Per_c$ is the perimeter

573: of the union of hexagons corresponding to $c$. Let $\ep$ be the largest

574: such constant. Let $\delta>0$ be such that the area of the Voronoi cell

575: in the fully densely loose-linked packing equals $1+\ep \Per_1 + \delta$,

576: where $\Per_1$ is the perimeter of the hexagon of a single tile. Since

577: $\ep = \rho + \O(\rho^2)$ and $\delta = \O(\rho^2)$, we have:

578: 

579: \begin{lem}\label{lem:voronoi}

580: For sufficiently small $\rho$, there are $\ep, \delta > 0$ such that for

581: any finite tight-linked component $c$,

582: \begin{eqnarray*}

583: d_1 &=&\frac{1}{1+ \Per_1 \ep + \delta}\,,\\

584: |V_c| &\ge& j_c+ \ep \Per_c\,, \\

585: \noalign{\hbox{and}}

586: \delta &\le& \ep/100\,.

587: \end{eqnarray*}

588: \end{lem}

589: 

590: \section{High Density}

591: Recall that $X$ is the compact space of all packings of the plane by the tile

592: (with the topology of Hausdorff-metric convergence on compact

593: subsets). Let ${\widetilde M}$ be the space of isometry-invariant Borel probability

594: measures on $X$. For any $\mu \in {\widetilde M}$, we denote by $|\mu|:=\mu(A_0)$

595: the {\bf density} of $\mu$, where $A_0$ is the set of all packings $P \in X$,

596: one of whose tiles contains the origin.  Since a tile is the closure of its interior,

597: $A_0$ is a closed set.

598: \begin{lem}\label{lem:cont}

599: If $\mu_i \in {\widetilde M}$ converges to $\mu$ in the weak* topology, then

600: $|\mu_i|$ converges to $|\mu|$.

601: \end{lem}

602: \begin{proof}

603: Let $\uP$ denote the union of tiles in a packing, $P$.

604: For any invariant probability measure $\nu$ and any $z \in \R^2$, we have

605: $$

606: |\nu| = \int \II{0 \in \uP} \,d\nu(P)

607: =

608: \int \II{z \in \uP} \,d\nu(P)

609: \,.

610: $$

611: Integrating over $z$ in a unit-area disk, $D$, with respect to Lebesgue

612: measure and using

613: Fubini's theorem gives the identity $|\nu| = \int |\uP \cap D| \,d\nu(P)$.

614: Since the function $P \mapsto |\uP \cap D|$ is continuous on $X$, the

615: lemma follows.

616: \end{proof}

617: 

618: Recall that $\lambda_0$ is the unique invariant measure on tilings 

619: by our tile, so that $|\lambda_0|=1$.  Recalling that $d_1$ is the density

620: of a fully densely

621: loose-linked tiling, fix a density $d$ with $d_1 \le d \le 1$. Let $\mu_N$

622: be the uniform measure on configuration of tiles at density $d_N$ in an $N \times N$

623: torus, where $d_N \to d$ as $N \to \infty$. To prove theorem

624: \ref{thm:main1}, we shall show that the weak* limit of $\mu_N$ exists and

625: equals $\cvxmu_s$, where $s:=\infrac{1-d}{ 1-d_1}$.

626: 

627: 

628: %\begin{cor}

629:   %The probability with respect to $\limmu_d$ that the origin is inside one of an infinite

630:   %connected component of linked tiles is nonzero for $d>d_1$.

631: %\end{cor}

632: 

633: We shall use several lemmas that depend on the following

634: notation.

635:  Given a packing $P \in X$, let

636: \begin{itemize}

637: \item $t_P$ be the tile of $P$ such that the origin belongs to $V(t_P)$

638: (this exists as long as the origin is not on the boundary of a Voronoi

639: cell),

640: \item $K_P$ be the tight-linked component containing $t_P$,

641: \item $j_P$ be the number of tiles in $K_P$, and

642: %\item $D_P$ be the disk of radius $1/100$ centered at the center of $t_P$,

643: \item $\f(P) := 3/j_P$ if $j_P$ is finite and $t_P$

644:   contains the origin, and 0 otherwise.

645: \end{itemize}

646: Thus $\f(P)$, in a sense, measures the number of degrees of freedom per tile near

647: the origin.

648: 

649: \begin{lem}\label{lem1} If $\nu$ is any measure in $\tM$, then $\int f\,d\nu(\f)

650:   \le 3|\nu|$, with equality iff $t_P$ has no tight links for $\nu$-almost

651: every packing $P$.

652: \end{lem}

653: \noindent

654: The proof is immediate.

655: %\vskip.2truein

656: %\noindent

657: %Example (not part of the proof): Let $X_N$ be the space of packings of

658: %an $N\times N$ 

659: %torus. Define $\f:~X_N \to \R$

660: %as above. Then $\frac{\mu_N(\f)}{d_N}$ is greater than or equal to the

661: %maximum number 

662: %of degrees of

663: %freedom per tile for a packing of the $N\times N$ torus at density $d_N$. 

664: 

665: \begin{lem}\label{lem:fcontinuity} If a sequence $\Seq{\nu_n} \subset \tM$ 

666: converges to $\nu$ in the weak* topology, then $\int f\,d\nu_n$

667: converges to $\int f\,d\nu$.

668: \end{lem}

669: 

670: \noindent

671: Lemma \ref{lem:fcontinuity} is proven in a manner similar to Lemma \ref{lem:cont}.

672: %\vskip.3truein

673: 

674: Given a finite tight-linked component $c$, let the congruence class of $c$

675: be $C$ and let

676: $X_C \subset X$ be the space of

677: all packings $P$ for which $t_P$ exists and

678: $K_P$ is in $C$. Let $X''$ be the

679: space of all packings $P$ with density $1$, where ``density'' refers to the

680: usual concept of the limit of the proportion of the area of $P$ inside a

681: large disk centered at the origin as the radius tends to infinity.  Let $X'

682: \subset X$ be the space of

683: all packings $P$ such that $K_P$ is infinite and either the density of $P$

684: is less than

685: 1, the density is not defined, or $t_P$ does not exist.

686: Thus, $X$ is the disjoint union of $X'$, $X''$ and 

687: the collection of $X_C$ for all $C$.

688: 

689: Let $\nu$ be any invariant probability measure with density $d$. Let $\nu_C$ be

690: $\nu$ conditioned on $X_C$, $\nu'$ be $\nu$ conditioned on $X'$ and $\nu''$ be

691: $\nu$ conditioned on $X''$. Since $\lambda_0$ is the only invariant probability

692: measure with support in $X''$, we have $\nu''=\lambda_0$. Thus,

693: $$

694: \nu = \nu(X')\nu' + \nu(X'')\lambda_0 + \sum_C \nu(X_C) \nu_C\,.

695: $$

696: 

697: Define the density $|\omega| := \omega(A_0)$ as before, but for any (invariant or non-invariant) probability

698: measure $\omega$ on $X$.

699: %$$

700: %\omega(\{P: \textrm{ there exists a tile

701:     %$t$ of $P$ 

702: %that contains the origin}\})\,.

703: %$$

704: %Note that when $\omega$ is invariant, this agrees with our prior

705: %definition.

706: We have

707: $$

708: d=|\nu| = \nu(X')\,|\nu'| + \nu(X'') + \sum_C  \nu(X_C)\,|\nu_C|

709: $$

710: and

711: $$

712: \int f\,d\nu = \sum_C \nu(X_C) \int f\,d\nu_C = \sum_C \nu(X_C)

713: \frac{3}{j_C}|\nu_C|\,,

714: $$

715: where $j_C$ is the number of tiles in $C$.

716: \begin{lem}\label{lem5}

717: Let $\nu \in \tM$ and $C$ be a finite-component class.

718: Suppose that $0 \le  s \le 1$ is such that 

719: $|\cvxmu_s| = |\nu_C|$. Then

720: $\int f\,d\cvxmu_s \ge \int f\,d\nu_C$. Moreover, equality holds only if $j_C = 1$. 

721: \end{lem}

722: 

723: \begin{proof}

724: As in the proof of Lemma \ref{lem:cont}, we have

725: that $|\nu_C| = \int j_C/|V(t_P)| \,d\nu_C(P)$.

726: 

727:   First suppose that $j_C=1$.  Then $|\nu_C|

728:   = \int 1/|V(t_P)| \,d\nu_C(P) \le d_1$ by Lemma \ref{lem:lambda1}.

729:   This means that $s=1$ and $\int f \,d\nu_C

730:   =3\,|\nu_C|=3\,|\cvxmu_s|= \int f \,d\cvxmu_s$. 

731:   

732:   Now assume that $j=j_C > 1$ and put $p :={\rm \Per}_C$. By definition,

733: $$

734: \int f\,d\cvxmu_s = s \int f \,d\lambda_1 + (1-s) \int f

735: \,d\lambda_0= s \int f \,d\lambda_1 = 3sd_1\,.

736: $$

737: Since $\nu_C(\f)=3|\nu_C|/j_C = 3|\cvxmu_s|/j_C = 3(sd_1 + 1-s)/j$, it suffices to show that

738: $$

739: sd_1 > \frac{sd_1 + (1-s)}{j}\,,

740: $$

741: which is equivalent to

742: $$

743: s (j d_1 - d_1 + 1) > 1 \,.

744: $$

745: Now $sd_1 + (1-s) = |\nu_C| \le \frac{j}{ j + \ep p}$, where $\ep$ is from

746: Lemma \ref{lem:voronoi}. Solving for $s$ gives

747: $$

748: s \ge \frac{1- \frac{j}{j+\ep p}}{ 1-d_1}

749: \,,

750: $$

751: whence it is enough to show that

752: $$

753: (jd_1 -d_1 + 1)\frac{1- \frac{j}{ j+\ep p}}{ 1-d_1} > 1\,.

754: $$

755: This boils down to

756: $$

757: d_1(p\ep+1) > 1\,.

758: $$

759: Now, $j>1$ implies that $p \ge (7/6)\Per_1$, where $\Per_1$ is the perimeter

760: of a single tile. Since $\ep/100 > \delta$ (by Lemma \ref{lem:voronoi}), this

761: implies that $p\ep + 1 > 1 + \ep \Per_1 + \delta = 1/d_1$, proving the last

762: inequality.

763: \end{proof}

764: 

765: \begin{lem}\label{lem6} We have

766: $\int f\,d\nu \le \int f\,d\cvxmu_s$ 

767: for all $\nu\in {\widetilde M}$ with $|\nu|=|\cvxmu_s|$.

768: Equality holds only if

769: \begin{itemize}

770: \item $\nu(X_C)=0$ for every component class $C$ with $j_C > 1$ and

771: \item whenever $\nu(X_C) > 0$ and $j_C = 1$, we have $|\nu_C|=d_1$.

772: \end{itemize}

773: \end{lem}

774: \begin{proof}

775: Recall that

776: $$

777: \int f\,d\nu = \sum_C \nu(X_C) \int f\,d\nu_C\,.

778: $$

779: For each component class $C$, let $s_C$ be defined as follows:

780: 

781: \begin{itemize}

782: \item if there exists $s \in [0,1]$ such that $|\nu_C| = sd_1 + (1-s)$, then

783: set $s_C:=s$;

784: \item otherwise, set $s_C:=1$.

785: \end{itemize}

786: Let $\omega_C := s_C\lambda_1 + (1-s_C)\lambda_0$ and

787: $$

788: \sigma := \big(\nu(X')+\nu(X'')\big)\lambda_0 + \sum_C \nu(X_C)\omega_C 

789: \,.

790: $$

791:  From the previous lemma, if $|\nu_C| \ge d_1$, then $\int \f \,d\nu_C

792: \le \int \f \,d\omega_C$,

793: with equality only if $j_C = 1$. If $|\nu_C| < d_1$, then $s_C = 1$ and

794: $$

795: \int \f \,d\nu_C = \frac{3\,|\nu_C|}{j_C} < 3d_1 = \int \f \,d\omega_C\,.

796: $$

797: 

798: Summing up, we obtain

799: \begin{eqnarray*}

800: \int \f \,d\sigma &=& \sum_C\nu(X_C) \int f\,d\omega_C\\

801: &\ge& \sum_C\nu(X_C)\int f\,d\nu_C=\int f\,d\nu\,.

802: \end{eqnarray*}

803: Moreover, equality holds only if $\nu(X_C)=0$ for every component $C$

804: with $j_C > 1$ and $|\nu_C|=d_1$ whenever $j_C = 1$.

805: Since $|\omega_C| \ge |\nu_C|$, we have

806: \begin{eqnarray*}

807: |\sigma| &=& \nu(X')+\nu(X'') + \sum_C \nu(X_C)\, |\omega_C|\\

808: &\ge& \nu(X')\,|\nu'|+\nu(X'') + \sum_C \nu(X_C)\, |\nu_C|\\

809: &=&|\nu|\\

810: &=&|\cvxmu_s|\,.

811: \end{eqnarray*}

812: Since $\sigma$ and $\cvxmu_s$ are both convex combinations of $\lambda_0$

813: and $\lambda_1$, this 

814: implies that $\int f\,d\sigma\le \int f\,d\cvxmu_s$ with equality iff

815: $\sigma=\cvxmu_s$. Thus, $\int f\,d\nu\le \int f\,d\cvxmu_s$. 

816: In the equality case we must have $\int f\,d\nu = \int f\,d\sigma=\int

817: f\,d\cvxmu_s$ and

818: $\sigma=\cvxmu_s$. This 

819: implies that $\nu(X_C)=0$ if $j_C > 1$ and

820: $|\nu_C|=d_1$ 

821: if $j_C = 1$.

822: \end{proof}

823: 

824: \begin{lem}\label{lem3}

825:   Let $\nu \in \tM$.

826:   If $|\nu|=|\cvxmu_s|$, then $\int f\,d\nu

827:   \le \int f\,d\cvxmu_s$. Equality holds iff $\nu=\cvxmu_s$.

828: \end{lem}

829: 

830: \noindent

831: (Informally, $\cvxmu_s$ uniquely maximizes the number of degrees of freedom per tile

832: for invariant measures of a fixed density.)

833: 

834: \begin{proof}

835:   The previous lemma implies $\int f\,d\nu\le \int f\,d\cvxmu_s$. Assume

836:   $\int f\,d\nu = \int f\,d\cvxmu_s$; then

837: $$

838: \nu = \nu(X')\nu' + \nu(X'')\lambda_0 + \nu(X_C)\nu_C

839: $$

840: where $C$ is the component of size 1 and $|\nu_C|=d_1$. 

841: This gives $\int f\,d\nu = \nu(X_C)

842: 3 d_1 = \int f\,d\cvxmu_s = 3s d_1$. Hence $\nu(X_C)=s$. Since $\nu'$ has density strictly

843: less than $1=|\lambda_0|$ but $|\nu|=|\cvxmu_s|$, we must have $\nu(X')=0$. That

844: is,

845: $$

846: \nu = \nu(X'')\lambda_0 + \nu(X_C)\nu_C

847: \,.

848: $$

849: Since $\nu$ and $\lambda_0$ are isometry invariant, $\nu_C$ must also be isometry 

850: invariant. By Lemma \ref{lem:lambda1}, $\lambda_1$ is the unique isometry-invariant

851: measure with support in $X_C$ and with density $d_1$. Hence $\nu_C = \lambda_1$. 

852: This implies $\nu=\cvxmu_s$ and the proof is finished.

853: \end{proof}

854: 

855: \medskip

856: 

857: \noindent

858: {\em Proof of Theorem~\ref{thm:main1}:}

859: It is easy to see that one can pack the $N\times N$ torus in such a way that there

860: is a large region of tight-linked tiles and a large region of densely loose-linked tiles,

861: in such a way that the interface between the two regions has a density which approaches

862: zero as $N$ tends to infinity, and the density $d_N$ of the packing $P_N$ tends to $d$.

863: Let $\omega_N$ be the invariant measure supported on isometric copies of ${\widetilde P_N}$

864: (a pull-back of $P_N$ to the plane). Then $\omega_N$ tends to $\cvxmu_s$ in the weak*

865: topology. By Lemma~\ref{lem:fcontinuity}, this implies that $\int

866: f\,d\omega_N \to \int f\,d\cvxmu_s$.

867:   

868: Now $\mu_N$, the uniform measure of density $d_N$ on the $N\times

869: N$ torus, satisfies $\int f\,d\mu_N \ge \int f\,d\omega_N$. This is because

870: $\mu_N$ is by definition supported on packings with the maximal

871: number of degrees of freedom for the given density $d_N$. Hence

872: $\liminf_N \int f\,d\mu_N \ge \liminf_N \int f\,d\omega_N = \int

873: f\,d\cvxmu_s$.

874:   

875: Therefore,

876: if $\mu_\infty$ is any weak* subsequential limit of $\Seq{\mu_N}_N$, then

877: $\int f\,d\mu_\infty \ge \int f\,d\cvxmu_s$. But $d_N \to d$, so $|\mu_\infty| =

878: |\cvxmu_s|$ by Lemma \ref{lem:cont}. The previous lemma now implies that

879: $\mu_\infty=\cvxmu_s$.  $\square$

880: 

881: Returning to the discussion of the introduction, we note that from

882: simulations of hard disks, one would expect the corollary to hold even

883: for a range of densities below $d_1$, but we do not know how to prove this.

884: 

885: \medskip

886: 

887: \noindent

888: {\bf Remark on Higher Dimensions}

889: 

890: \smallskip

891: 

892: The basic features of our argument can be generalized to dimension 3 or

893: higher, except for our use in Lemma \ref{lem:lambda1} of \cite{Fej2} on

894: the minimal Voronoi region in

895: disk packings in the plane. It would be of interest if this part of

896: our proof could be replaced by an argument insensitive to

897: dimension.

898: 

899: \section{Low Density}

900: 

901: In this final section, we confirm the intuition that at low densities,

902: there will be no infinite loose-linked component.  It is obvious that there

903: is no infinite {\em tight}-linked component at densities smaller than $d_1$.

904: 

905: We begin with

906: a lemma that holds for any tile shape (in fact, for any {\em collection}

907: of shapes and sizes, as long as each can be fit into a disk of some fixed

908: radius $s$, and ``density'' is interpreted as number of tiles per unit area).

909: 

910: \begin{lem}\label{lem:poisson} For small enough density $d$,

911: if a packing $P$ is drawn from $\limmu_d$, then the probability that the

912: disk $B_R$ of radius $R$ about the origin contains more than $9R^2d$ tile centers

913: goes to zero as $R \to \infty$.

914: \end{lem}

915: 

916: \begin{proof} Let $s$ be the radius of the smallest disk containing the tile

917: (in our case, $s$ is about $2^{1/2} \cdot 3^{-3/4} \cdot (1\!+\!2\rho)$) and choose

918: $$

919: 0 < d < \frac{.05}{13 \pi s^2}~;

920: $$

921: for our zipper tiles with small enough $\rho$, $d \le .003$ suffices.

922: Let $T$ be the set of tiles whose centers fall in $B_R$, $k := \lceil \pi

923: R^2 d\rceil$

924: and $\ell > 9 R^2 d$.  Letting $\limmu_d(\cdot)$ denote

925: the probability of an event with respect to the measure $\limmu_d$,

926: we shall show that

927: $$

928: \frac{\limmu_d(|T|\!=\!\ell)}{\limmu_d(|T|\!=\!k)} \le \gamma^{\ell-k}

929: $$

930: for some constant $\gamma < 1$.  It then follows that

931: $$

932: \limmu_d(|T|> 9 R^2 d) \le {\limmu_d(|T|\!=\!k)}\sum_{\ell=\lceil 9R^2d \rceil}^\infty

933: \gamma^{\ell-k} \le

934: {\limmu_d(|T|\!=\!k)}\frac{\gamma^{\lfloor (9-\pi)R^2d \rfloor}}{1\!-\!\gamma}\to 0

935: $$

936: as $R \to \infty$, as desired.

937: 

938: The measure $\limmu_d$ is the limit of uniform distributions of

939: configurations on the $N \times N$ torus ${\bf T}_N$, in turn obtainable by

940: choosing a sequence of $n = \lfloor N^2 d \rfloor$ points from the

941: Lebesgue distribution $\lambda$ on ${\bf T}_N^n$ as the centers of the tiles,

942: orienting each tile independently and uniformly at random, and finally conditioning

943: on no overlap.  We denote by $\lambda(|T|\!=\!j)$ the

944: {\em a priori} probability that exactly $j$ points fall inside $B_R$ (which

945: we take to be some fixed disk in the torus).

946: 

947: Let $\Phi$ be the event that there is no overlap among the tiles whose centers

948: lie in $B_R$, and $\Psi$ the event that there is no overlap involving any tile

949: whose center falls {\em outside} $B_R$.  Then

950: $$

951: \frac{\limmu_d(|T|\!=\!\ell)}{\limmu_d(|T|\!=\!k)} = \frac{\lambda(|T|\!=\!\ell)}{\lambda(|T|\!=\!k)}

952: \cdot \frac{\lambda(\Phi\bigm||T|\!=\!\ell)}{\lambda(\Phi\bigm||T|\!=\!k)} \cdot

953: \frac{\lambda(\Psi\bigm||T|\!=\!\ell\,\wedge\,\Phi)}{\lambda(\Psi\bigm||T|\!=\!k\,\wedge\,\Phi)}~.

954: $$

955: and our job is to bound the three fractions on the right.

956: 

957: For the first, we note that $|T|$ is binomially distributed in the measure $\lambda$,

958: hence

959: \begin{eqnarray*}

960: \frac{\lambda(|T|\!=\!\ell)}{\lambda(|T|\!=\!k)} &=& \frac{ {n \choose \ell} \left(\frac{\pi R^2}{N^2} \right)^\ell

961: \left(1 - \frac{\pi R^2}{N^2} \right)^{n-\ell} }{ {n \choose k} \left(\frac{\pi R^2}{N^2} \right)^k

962: \left(1 - \frac{\pi R^2}{N^2}\right)^{n-k} }

963: \le\frac{(n\!-\!k)!/(n\!-\!\ell)!}{\ell!/k!} \left( \frac{

964: \frac{k}{n} }{ 1-\frac{k}{n} } \right)^{\ell-k}\\

965: &<& \frac{ (n\!-\!k)^{\ell-k} } { (\ell/e)^{\ell-k} } \cdot \frac { k^{\ell-k} } { (n\!-\!k)^{\ell-k} }

966: = \left(\frac{ek}{\ell} \right)^{\ell-k}

967: \le 0.95^{\ell-k}

968: \end{eqnarray*}

969: for large $R$.

970: 

971: The next fraction is easy:  since we may throw the first $k$ centers into $B_R$, then

972: the remaining $\ell\!-\!k$, we have

973: $$

974: \frac{\lambda(\Phi\bigm||T|\!=\!\ell)}{\lambda(\Phi\bigm||T|\!=\!k)}

975: $$

976: is the probability that the additional $\ell\!-\!k$ centers do not cause a collision, which is at most 1.

977: 

978: For the (inverse of) the third fraction, we throw $n\!-\!\ell$ centers into the region outside $B_R$,

979: then the remaining $\ell\!-\!k$.  A new point, if it lands at distance greater than $2s$ from

980: any previous point or from the disk $B_R$, causes no new overlap; and at each stage there

981: are fewer than $n\!-\!k$ points already placed.  Hence

982: $$

983: \frac{\lambda(\Psi\bigm||T|\!=\!k\,\wedge\,\Phi)}{\lambda(\Psi\bigm||T|\!=\!\ell\,\wedge\,\Phi)}

984: > \left( \frac{N^2 - \pi (R + 2s)^2 - (n\!-\!k)4 \pi s^2}{N^2 - \pi R^2} \right)^{\ell-k}

985: $$

986: $$

987: \ge \left(1 - 4 \pi s^2 d - \frac{4\pi sR + 4\pi s^2}{N^2 - \pi R^2}\right)^{\ell-k} 

988: > \left(1 - 13 s^2 d\right)^{\ell-k}

989: $$

990: for $N \gg R$.

991: 

992: Putting the inequalities together, we have

993: $$

994: \frac{\limmu_d(|T|\!=\!\ell)}{\limmu_d(|T|\!=\!k)}

995: \le \left(\frac{.95}{1-13 s^2 d}\right)^{\ell-k} = \gamma^{\ell-k}

996: $$

997: where $\gamma := .95/(1-13s^2 d) < 1$ by choice of $d$.

998: \end{proof}

999: 

1000: \begin{pro}\label{thm:low} For some $d_2>0$, the $\limmu_d$-probability that the origin

1001: is inside an infinite connected component of loosely-linked tiles is zero for $d<d_2$.

1002: \end{pro}

1003: 

1004: \begin{proof}

1005: Let $d \in (0,.003)$ be a density to be chosen later. Let $P$ be a

1006: packing drawn from $\limmu_d$; we aim to show that the probability that

1007: the origin is connected by a loose-linked chain of tiles of $P$ to

1008: some point at distance $R$ approaches zero as $R \to \infty$.

1009:   

1010: We again choose some large radius $R$ and let $T$ be the set of tiles of $P$

1011: whose centers fall inside the disk $B_R$. 

1012: 

1013: Fix the positions of the tiles of $P \setminus T$ (the black tiles of Figure~\ref{fig:chain})

1014: and consider the space of packings having these tiles plus $n$ tiles whose centers

1015: fall in $B_R$. We think of this space as being a subset of $T_1(B_R)^n/\Sym(n)$,

1016: where $T_1(B_R)$ is the unit tangent bundle of $B_R$ (modulo a $2\pi/6$ rotation to

1017: take into account the symmetries of the tile) and the symmetric group

1018: acts by permuting the factors.

1019: 

1020: If $\alpha_n$ is the volume (in $T_1(B_R)^n/\Sym(n)$-space) of this

1021: space and $m < n$, then by packing $n\!-\!m$ tiles into $B_R$ and then

1022: the remaining $m$ in the left-over space, we have

1023: $$

1024: \alpha_n \ge \frac{1}{{n \choose {n-m}}}\alpha_{n-m} \frac{1}{m!}

1025: \big[\pi(R-2s)^2 - n\pi(2s)^2\big]^m

1026: \,,

1027: $$

1028: where $s$ is, as before, the radius of the circle circumscribing a tile.  This

1029: takes into account possible intrusion of tiles in $P-T$ into $B_R$,

1030: and the fact that a tile center at point $x$ can exclude nearby

1031: centers but only within distance $2s$ of $x$.

1032: 

1033: \begin{figure}[htp]

1034: \epsfxsize=3.5truein

1035: $$\epsfbox{chain.ps}$$

1036: %\centerline{\epsfig{figure=hex7.ps}}

1037: \caption{An unlikely configuration of tiles in and around $B_R$}\label{fig:chain}

1038: \end{figure}

1039: 

1040: Let $\beta$ denote the ``wiggle room'' of a tile $t$ loose linked to a

1041: stationary tile $t'$, that is, the 3-dimensional volume of the space

1042: of positions of $t$; then $\beta = \O(\rho^3)$ (but we use only

1043: that $\beta$ is bounded by a constant).  If a packing ``percolates'',

1044: that is, contains a chain of loose-linked tiles connecting the center

1045: to the boundary of $B_R$, let $t_1, \dots, t_m$ be a shortest such

1046: chain (the dark grey tiles of Figure~\ref{fig:chain}). Note that

1047: $m \ge R/(2s)$.  For each $i>2$, the tile $t_i$ is linked to one of the

1048: three sides of $t_{i-1}$ farthest from the side of $t_{i-1}$ linked to

1049: $t_{i-2}$, and has wiggle room at most $\beta$ with respect to

1050: $t_{i-1}$.  Accounting for the orientation of $t_1$ and allowing the

1051: remaining $n\!-\!m$ tile centers to fall anywhere in $B_R$, we

1052: have that the $3n$-dimensional volume of the set of percolating packings

1053: is bounded by $(\pi/3)\cdot 6 \cdot 3^{m-2} \cdot \beta^{m-1} \cdot \alpha_{n-m}

1054: < 3^m \beta^{m-1} \alpha_{n-m}$.

1055: 

1056: Comparing with the lower bound for $\alpha_n$, we find that given

1057: $|T|=n \le 9dR^2$, the probability of percolation is less than

1058: $$

1059: \frac{3^m \beta^{m-1} \alpha_{n-m}}{ \alpha_n}

1060: \le \frac{3^m \beta^{m-1} \, n!/(n\!-\!m)!}{\big[\pi(R-2s)^2 - n\pi(2s)^2\big]^m}

1061: < \Big(\frac{27 \beta d R^2}{\pi\big[(R-2s)^2 - 36 d R^2 {s}^2\big]}\Big)^m /\beta

1062: \,,

1063: $$

1064: which goes to zero as $R$ (thus also $m$) increases, for suitably chosen $d$.

1065: Since we know from Lemma~\ref{lem:poisson} that $\limmu_d(|T| \le 9dR^2)$ approaches

1066: 1 as $R \to \infty$, the proposition follows.

1067: \end{proof}

1068: 

1069: A more careful argument would prove Proposition~\ref{thm:low} for any density

1070: below $1/\big(4\pi(2/3\sqrt{3})\big) = .2067^+$ for sufficiently small $\rho$, but clearly the probability of percolation

1071: will remain 0 for much higher densities than that.

1072: 

1073: \section{A Conjecture}

1074: 

1075: We have shown that high-density random packings of zipper tiles in the plane contain

1076: an infinite loose-linked component with positive probability, while

1077: low-density random 

1078: packings do not.

1079: What happens in the case of ordinary disks, where there is no apparent linking

1080: mechanism?  We believe, but cannot prove, the following

1081: 

1082: \medskip

1083: 

1084: \noindent

1085: {\bf Conjecture.}  {\em Suppose $\limmu_d$ is defined as above

1086: for geometric disks of radius 1.  Join two centers by an edge 

1087: if their distance is at most $2 + \ep$ for some fixed $\ep \ll 1$. 

1088: Then for sufficiently high density $d$ below the maximum, the graph

1089: resulting from a configuration drawn from $\limmu_d$ will contain an

1090: infinite connected component a.s.}

1091: 

1092: \smallskip

1093: This statement can be shown by a standard Peierls-type argument for large

1094: $\ep$; this may be known already.

1095: In general, there is some parameter set of $(\ep, d) \subset (0, \infty)

1096: \times (0, 1)$ for which there is an infinite component.

1097: For small $d$ or for large $\ep$, the problem is quite similar to continuum

1098: percolation, where one connects by an edge two points of a Poisson point

1099: process if their distance is at most $r$.

1100: Because of homotheties, one may fix the

1101: intensity of the point process to be 1. Then there is a phase transition in

1102: $r$. Our problem is quite different in really having two parameters,

1103: but our conjecture is that there is a phase transition in $d$ for every

1104: $\ep$ nevertheless.

1105: 

1106: 

1107: \begin{thebibliography}{xxx}

1108:   

1109: \bibitem[AH]{AH} B. J. Alder, W. G. Hoover, Numerical Statistical

1110:   Mechanics, in {\it Physics of Simple Liquids}, edited by

1111: H. N. V. Temperley, J. S. Rowlinson and G. S.

1112: Rushbrooke, (John Wiley, New York, 1968) 79--113.

1113: 

1114: \bibitem[BLRW]{BLRW} L. Bowen, R. Lyons, C. Radin and P. Winkler,

1115:   Fluid/solid transition in a hard-core system, to appear in

1116:   {\it Phys. Rev. Lett.} Preprint available

1117:   from \texttt{http://www.arxiv.org/abs/cond-mat/0508405}.

1118:   

1119: \bibitem[BHRS]{BHRS} L. Bowen, C. Holton, C. Radin and L. Sadun,

1120: Uniqueness and symmetry in problems of optimally dense packings,

1121: {\it Math. Phys. Electron. J.} 11(2005), Paper 1.

1122: 

1123: \bibitem[DTSC]{DTSC} A. Donev, S. Torquato, F. H. Stillinger and R. Connelly, Jamming in

1124:   hard sphere and disk packings, {\it J. Appl. Phys.} {\bf 95} (2004), 989--999.

1125: 

1126: \bibitem[Fej]{Fej} L. Fejes T\'oth, {\it Regular Figures},

1127:   Macmillan, New York, 1964.

1128: 

1129: \bibitem[Fej2]{Fej2} L. Fejes T\'oth, {\it Lagerungen in der Ebene,

1130:     auf der Kugel und im Raum}, 2nd Ed., Springer, Berlin, 1972, pp. 62--63.

1131: 

1132: \bibitem[Gri]{G} G. Grimmett, {\em Percolation}, 2nd Ed., Springer-Verlag,

1133:   Berlin 1999.

1134: 

1135: \bibitem[KRS]{KRS} H. Koch, L. Sadun and C. Radin, Most stable

1136:   structure for hard spheres, {\it Phys. Rev. E} {\bf 72} (2005), paper 016708.

1137: 

1138: \bibitem[Rad]{Rad} C. Radin, Orbits of orbs: sphere packing meets

1139:   Penrose tilings, {\it Amer. Math. Monthly} {\bf 111} (2004), 137--149.

1140: 

1141: \end{thebibliography}

1142: 

1143: \end{document}

1144: