1: \date{} 

2: \title{Quantitative noise sensitivity\\and exceptional times for percolation}

3: \author{Oded Schramm \and Jeffrey E. Steif

4: \thanks{Research supported by the

5:      Swedish Natural Science Research Council

6: and the G\"{o}ran Gustafsson Foundation (KVA).}}

7: 

8: 

9: \documentclass[12pt,naturalnames]{article}

10: \usepackage{amsmath}

11: \usepackage{amssymb}

12: \usepackage{amsthm}

13: \usepackage{amsfonts}

14: \usepackage{graphicx}

15: \input labelfig.tex

16: \input epsf.tex

17: 

18: \newif\ifhyper\IfFileExists{hyperref.sty}{\hypertrue}{\hyperfalse}

19: 

20: \ifhyper\usepackage{hyperref}\fi

21: 

22: \newif\ifdraft

23: \drafttrue

24: \def\note#1{\ifdraft {\bf [#1]}\fi}

25: \long\def\comment#1{}

26: \long\def\oldVersion#1{}

27: \def\dhrule{\bigskip\hrule\smallskip\hrule\bigskip}

28: \numberwithin{equation}{section}

29: \numberwithin{figure}{section}

30: 

31: \newtheorem{theorem}{Theorem}

32: \numberwithin{theorem}{section}

33: \newtheorem{corollary}[theorem]{Corollary}

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

35: \newtheorem{proposition}[theorem]{Proposition}

36: \newtheorem{conjecture}[theorem]{Conjecture}

37: \newtheorem{problem}[theorem]{Problem}

38: \newtheorem{exercise}[theorem]{Exercise}

39: \theoremstyle{remark}\newtheorem{definition}[theorem]{Definition}

40: \theoremstyle{remark}\newtheorem{remark}[theorem]{Remark}

41: \def\eref#1{(\ref{#1})}

42: \def\QED{\qed\medskip}

43: \newcommand{\Prob} {{\bf P}}

44: \newcommand{\R}{\mathbb{R}}

45: \newcommand{\C}{\mathbb{C}}

46: \newcommand{\Z}{\mathbb{Z}}

47: \newcommand{\N}{\mathbb{N}}

48: \def\diam{\mathop{\mathrm{diam}}}

49: \def\area{\mathop{\mathrm{area}}}

50: \def\dist{\mathop{\mathrm{dist}}}

51: \def\length{\mathop{\mathrm{length}}}

52: \def\ceil#1{\lceil{#1}\rceil}

53: \def\floor#1{\lfloor{#1}\rfloor}

54: \def\Im{{\rm Im}\,}

55: \def\calC{{\cal C}}

56: \def\calN{{\cal N}}

57: \def\bPsi{\mbox{\boldmath $\Psi$}} 

58: \def\Re{{\rm Re}\,}

59: \def\FC(#1,#2){{ 0 \stackrel{#2}{\leftrightarrow} #1 }}

60: \def\Ito/{It\^o}

61: \def \eps {\epsilon}

62: \def \P {{\bf P}}

63: \def\md{\mid}

64: \def\Bb#1#2{{\def\md{\bigm| }#1\bigl[#2\bigr]}}

65: \def\BB#1#2{{\def\md{\Bigm| }#1\Bigl[#2\Bigr]}}

66: \def\Bs#1#2{{\def\md{\mid}#1[#2]}}

67: \def\Pb{\Bb\P}

68: \def\Eb{\Bb\E}

69: \def\PB{\BB\P}

70: \def\EB{\BB\E}

71: \def\Ps{\Bs\P}

72: \def\Es{\Bs\E}

73: \def \prob {{\bf P}}

74: \def \expect {{\bf E}}

75: \def \p {{\partial}}

76: \def \E {{\bf E}}

77: \def\defeq{:=}

78: \def\var{\operatorname{var}}

79: \def\closure{\overline}

80: \def\ev#1{{\mathcal{#1}}}

81: \def \proof {{ \medbreak \noindent {\bf Proof.} }}

82: \def\proofof#1{{ \medbreak \noindent {\bf Proof of #1.} }}

83: \def\proofcont#1{{ \medbreak \noindent {\bf Proof of #1, continued.} }}

84: \def\CCa{C_3}

85: \def\CCb{C_1}

86: \def\CCc{C_2}

87: 

88: \def\bl{\bigl}\def\br{\bigr}\def\Bl{\Bigl}\def\Br{\Bigr}

89: \def\Vbdry{V_{\p}}

90: \def\V#1{V(#1)}

91: \def\Vss{V_{s,R}\cap V_{s',R}}

92: \def\Vb#1{V_\p(#1)}

93: \def\VD{V_D}

94: \def\sign{\mathrm{sign}}

95: 

96: \def\nn{[n]}

97: \def\cro{Q}

98: \def\croRr{f_r^R}

99: \def\mfh{\mathfrak H}

100: \def\hmfh{\hat\mfh}

101: \def\interior{\operatorname{interior}}

102: \def\II{{J}}

103: %

104: 

105: \def\noopsort#1{}

106: 

107: \hfuzz 1pt

108: %

109: \begin{document}

110: \maketitle

111: 

112:  \begin{abstract}

113: One goal of this paper is to

114:  prove that dynamical critical site percolation on the planar triangular lattice

115: has exceptional times at which percolation occurs. 

116: In doing so, new {\sl quantitative} noise sensitivity results

117: for percolation are obtained.

118: The latter is based on a novel method for controlling the

119:  ``level $k$'' Fourier coefficients via the construction of a

120: randomized 

121: algorithm which looks at random bits, outputs the value of a 

122: particular function but looks at any fixed input bit with low 

123: probability. We also obtain upper and lower bounds on the Hausdorff

124: dimension of the set of percolating times. We then study the problem

125: of exceptional times for certain

126: ``$k$-arm'' events on wedges and cones. As a corollary

127: of this analysis, we prove, among other things, that there are no times 

128: at which there are two infinite ``white'' clusters, obtain an upper bound 

129: on the Hausdorff dimension of the set of times at which there are 

130: both an infinite white cluster and an infinite black cluster

131: and prove that for dynamical critical bond percolation on the square grid

132: there are no exceptional times at which $3$ disjoint infinite clusters are

133: present.

134: \end{abstract}

135: 

136: \newpage

137: \tableofcontents

138: \newpage

139: 

140: \section {Introduction}

141: 

142: Consider bond percolation on an infinite connected locally finite graph

143: $G$, where for some $p\in[0,1]$, each edge (bond) of $G$ is, independently of 

144: all others, open with probability $p$ and closed with probability $1-p$. 

145: Write $\pi_p$ for this product measure. 

146: The main questions in percolation theory (see \cite{Grimmett})

147: deal with the possible existence of infinite connected components 

148: (clusters) in the random subgraph of $G$

149: consisting of all sites and all open edges. 

150: Write $\calC$ for the event

151: that there exists such an infinite cluster. By Kolmogorov's 0-1 law, 

152: the probability of $\calC$ is, for fixed $G$ and $p$, either 0 or 1. 

153: Since $\pi_p(\calC)$ is nondecreasing in $p$, there 

154: exists a critical probability $p_c=p_c(G)\in[0,1]$ such that

155: \[

156: \pi_p(\calC)=\left\{

157: \begin{array}{ll}

158: 0 & \mbox{for } p<p_c \\

159: 1 & \mbox{for } p>p_c.

160: \end{array} \right. 

161: \]

162: At $p=p_c$ we can have either $\pi_p(\calC)=0$ or $\pi_p(\calC)=1$, depending on $G$.

163: 

164: 

165: H\"{a}ggstr\"{o}m, Peres and Steif~\cite{HPS} initiated the study of 

166: dynamical percolation.

167: (The notion of dynamical percolation

168: was invented independently by I.~Benjamini.

169: While the present paper was motivated by \cite{HPS},

170: the question studied here had previously been asked by Benjamini, as

171: we recently became aware.)

172: In this model, with $p$ fixed, the edges of $G$ switch back 

173: and forth according to independent 2 state

174: continuous time Markov chains where closed switches to open at rate $p$ 

175: and open switches to closed at rate $1-p$. Clearly $\pi_p$ is a stationary

176: distribution for this Markov process. The general question studied in

177: \cite{HPS} was whether, when we start with distribution $\pi_p$,

178: there could exist atypical times at which the percolation 

179: structure looks markedly different than at a fixed time.

180: 

181: Write $\bPsi_p$ for the underlying probability measure

182: of this Markov process, and write $\calC_t$ for

183: the event that there is an infinite cluster of open edges 

184: at time $t$.

185: 

186: Two results in \cite{HPS} which are relevant to us are

187: 

188: \begin{proposition} \label{pr:noncrit}

189:  For any graph $G$ we have 

190: \begin{equation} \left\{ \begin{array}{ccl}

191: \bPsi_p(\, \calC_t \, \mbox{ occurs for every  } \, t \, )=1  & \mbox{ if } & p>p_c(G) 

192:    \\[1ex]

193: \bPsi_p\bigl((\neg\, \calC_t) \mbox{ occurs for every } t\bigr)=1 & \mbox{ if } & p<p_c(G) \, .

194: \end{array} \right.

195: \nonumber

196: \end{equation} 

197: \end{proposition}

198: 

199: \begin{theorem} \label{th:Zd}

200: For $d\ge 19,$ the integer lattice $\Z^d$ satisfies

201: $$

202: \bPsi_{p_c}\bl((\neg \calC_t) \mbox{ occurs for every } t\br)=1.

203: $$ 

204: \end{theorem}

205: 

206: One important aspect of the proof of the latter result is that it uses the fact,

207: proved in \cite{HS},

208: that for $d\ge 19$, 

209: \begin{equation}\label{e.z19}

210: \pi_p(0\text{ is in an infinite open cluster})

211: =O(|p-p_c|).

212: \end{equation}

213: It is proved  in \cite{KZ} that~\eref{e.z19} does not hold

214: for $d=2$.

215: Therefore, the question of whether Theorem 

216: \ref{th:Zd} is true for $d=2$ becomes interesting. At this point, we mention

217: that site percolation is the analogous model where the vertices (rather than the edges)

218: are open or closed independently each with probability $p$ and

219: dynamical percolation is defined in a completely analogous manner.

220: Our main result says that Theorem~\ref{th:Zd} does not hold

221: for site percolation on  the planar triangular grid.

222: The triangular grid is the graph whose vertex set

223: is the subset of $\C=\R^2$ consisting of the points

224: $$

225: \Z+\exp(2\,\pi\,i/3)\,\Z=

226: \bigl\{(k+\ell/2,\sqrt 3\, \ell/2):k,\ell\in\Z\bigr\}

227: $$

228: and two such points have an edge between them if and only if their distance is 1.

229: Explicitly stated, our main result is

230: 

231: 

232: \begin{theorem} \label{th:except}

233: Almost surely, the set of times

234: $t\in[0,1]$ such that dynamical critical site percolation

235: on the triangular lattice has an infinite open cluster is nonempty.

236: \end{theorem}

237: 

238: There are no other transitive graphs for which it is known that

239: dynamical critical percolation has such exceptional times.

240: (In~\cite{HPS}, it was argued that the event discussed in Theorem~\ref{th:except}

241: is measurable. A similar comment applies to our other results below.

242: Thus, measurability issues

243: will not concern us here.)

244: 

245: We are convinced that Theorem~\ref{th:except}

246: is true for bond percolation on the square lattice. However, our proof uses 

247: the existence and exact values of certain so-called 

248: {\sl critical exponents}, which

249: are only known to hold for site percolation on the triangular lattice.

250: These are believed to be the same for 

251: bond percolation on the square lattice, but even

252: their existence has not yet been established in that case. 

253: However, the methods of this paper seem to come quite close

254: to a proof for the square grid as well: it seems that there are several

255: ways in which this can perhaps be achieved without determining these

256: critical exponents. These issues will be further discussed in Section~\ref{sec:square}.

257: 

258: It is interesting to note that by~\cite[Corollary 4.2]{HPS}, a.s.\ at every

259: time $t$ the set of vertices that are contained in some infinite cluster

260: has zero density.

261: \bigskip

262: 

263: On a heuristic level, for Theorem~\ref{th:except} to hold,

264: it is necessary that

265: the configuration ``changes fast'' in order to have ``many chances''

266: to percolate  so that we will in fact have a percolating time. 

267: Mathematically, ``changing fast'' can be interpreted as having small

268: correlations over short time intervals, which then suggests the

269: use of the second moment method which we indeed will use. In other words,

270: one needs to know that the configuration at a given time tells us almost

271: nothing about how it will look a short time later. The notion of

272: ``noise sensitivity'' introduced in \cite{BKS} 

273: is the relevant tool which

274: describes this phenomenon. We now briefly explain this.

275: 

276: Given an integer $m$, a subset $A$ of $\{0,1\}^m$ and an $\eps>0$, define

277: $$

278: N(A,\eps):=

279: \var\Bigl[\Pb{(Y_1,\ldots,Y_{m})\in A\md X_1,\ldots,X_{m}}\Bigr]

280: $$

281: where 

282: $\{X_i\}_{1\le i\le m}$ are i.i.d.\ with

283: $\Pb{X_i=1}=1/2=\Pb{X_i=0}$ and conditional on the $\{X_i\}$'s,

284: $\{Y_i\}_{1\le i\le m}$ are independent with

285: $Y_i=X_i$ with probability $1-\epsilon$ and

286: $Y_i=1-X_i$ with probability $\epsilon$. 

287: 

288: \begin{definition} \label{def:noise}

289: Let $\{n_m\}_{m\ge 1}$ be an increasing sequence in $\N$ going to $\infty$ and 

290: let $A_m$ be a subset of $\{0,1\}^{n_m}$ for each $m$. We say that

291: the sequence $\{A_m\}_{m\ge 1}$ is {\bf noise sensitive} if for every $\epsilon >0$,

292: \begin{equation}\label{e.noisedef}

293: \lim_{m\to\infty} N(A_m,\eps)=0.

294: \end{equation}

295: \end{definition}

296: 

297: This says that for large $m$ knowing the values of

298: $X_1,\ldots,X_{n_m}$ gives us almost no information concerning whether

299: $(Y_1,\ldots,Y_{n_m})\in A_m$. This is not the exact definition of noise sensitivity

300: given in \cite{BKS} 

301: but is easily shown to be equivalent; see page 14 in that paper. It is also

302: shown in \cite{BKS} that if (\ref{e.noisedef}) holds for some 

303: $\epsilon \in(0, 1/2)$, then it holds for all such $\epsilon$

304: and in addition that $N(A,\eps)$ is decreasing in $\epsilon$

305: on $[0,1/2]$.

306: 

307: Let $n_m$ be the number of edges in an $(m+1)\times m$ box in $\Z^2$ and let $A_m$ be the event

308: of a left to right crossing in such a box.

309: By duality, $\Pb{A_m}=1/2$ for every $m$

310: (see~\cite{Grimmett}).

311: %(see \cite{Grimmett}, page 294). 

312: In \cite{BKS}, the following result is proved.

313: 

314: \begin{theorem} \label{th:crossing}

315: The sequence $\{A_m\}_{m\ge 1}$ is noise sensitive.

316: \end{theorem}

317: 

318: A by-product of the tools needed to prove

319:  Theorem \ref{th:except} will imply the following more quantitative

320: version of Theorem~\ref{th:crossing}, which was conjectured in~\cite{BKS}.

321: 

322: \begin{theorem} \label{th:crossingquantsquare}

323: There exists $\gamma >0$ so that

324: $$

325: \lim_{m\to\infty} N(A_m,m^{-\gamma})=0.

326: $$

327: \end{theorem}

328: 

329: 

330: We have the same result for the triangular lattice but with a better $\gamma$, since

331: critical exponents are known in this case.

332: 

333: \begin{theorem} \label{th:crossingquant} For critical

334: site percolation on the triangular lattice,

335: let $A_m'$ be the event of the existence of a left-right crossing

336: in a domain $D$ approximating a square of sidelength $m$.

337: Then for all $\gamma < 1/8$,

338: $$

339: \lim_{m\to\infty} N(A_m',m^{-\gamma})=0.

340: $$

341: \end{theorem}

342: 

343: In proving our quantitative noise sensitivity results

344: (Theorems \ref{th:crossingquantsquare} and \ref{th:crossingquant}

345: as well as those later on necessary for obtaining Theorem \ref{th:except}),

346: one of two key steps will be Theorem \ref{t.noise}, which gives 

347: estimates of certain quantities involving

348: Fourier coefficients of a function based on the properties of an algorithm

349: calculating the function;

350: the other key step will be the construction of an appropriate algorithm.

351:  Precise definitions of undefined terms 

352: will be given in Section \ref{sec:dilute},

353: where the connection with noise sensitivity will also be recalled. 

354: 

355: \begin{theorem}\label{t.noise}

356: Let $n\in\N$ and set $\Omega=\Omega_n:=\{0,1\}^n$.

357: Let $f:\Omega\to\R$ be a function.

358: Suppose that there is a randomized algorithm $A$ for

359: determining the value of $f$ which examines some of the input bits of

360: $f$ one by one, where the choice of the next bit examined may depend

361: on the bits examined so far. Let $\II\subseteq \nn:=\{1,2,\dots,n\}$

362: be the (random) set of bits examined by the algorithm.

363: Set $\delta=\delta_A:=\sup\bl\{\Ps{i\in \II}:i\in\nn\br\}$.

364: Then, for every $k=1,2,\dots$,

365: the Fourier coefficients of $f$ satisfy

366: \begin{equation}\label{e.noise}

367: \sum_{S\subseteq\nn,\,|S|=k}\hat f(S)^2 \le \delta\,k\,\|f\|^2,

368: \end{equation}

369: where $\|f\|$ denotes the $L^2$ norm of $f$ with respect to the

370: uniform probability measure on $\Omega$.

371: \end{theorem}

372: 

373: This result might have some applications to 

374: theoretical computer science. We will call $\delta_A$

375: the {\bf revealment} of the algorithm $A$. 

376: The restriction of $x$ to $J$ (the set of bits examined by the algorithm)

377: is a {\bf witness} for the function $f$, in the sense that it determines

378: $f(x)$. As explained in Section \ref{ss.noisetheorem}, Theorem~\ref{t.noise} extends to some

379: other types of witnesses.

380: 

381: In the case $k=1$, the inequality~\eref{e.noise} cannot be improved

382: by more than a factor of $O(1/\log n)$:

383: there is an example showing this with $\delta\le n^{-1/3}\,\log(n)$,

384: which appears in~\cite[\S 4]{BSW}.

385: The paper~\cite{BSW} investigates how small the revealment can be for

386: a balanced boolean function on $\{0,1\}^n$. When the function is

387: monotone, it is shown that the revealment cannot be much smaller than

388: $n^{-1/3}$ and in general it cannot be much smaller than

389: $n^{-1/2}$. Examples are given there which come within logarithmic

390: factors of meeting these bounds.

391: 

392: We don't know if~\eref{e.noise} is close to being optimal for $k\gg 1$.

393: One is tempted to speculate that the inequality can be improved

394: to $\sum_{|S|\le k}\hat f(S)^2\le O(1)\,k\,\delta\,\|f\|^2$.

395: We do not know any counterexample to this inequality.

396: However, the AND function $f(x)=\prod_{j=1}^n x_j$

397: gives an example where

398: $$

399: O(1)\,\sum_{|S|= k}\hat f(S)^2 \ge\sqrt k\, \delta\,\|f\|^2

400: $$

401: for $k$ satisfying $|k-n/2|=O(n^{1/2})$.

402: (It is easy to check that the best revealment 

403: possible for this $f$ is exactly $(2-2^{1-n})/n$.)

404: 

405: \bigskip

406: 

407: 

408: Once Theorem \ref{th:except} is established, it is natural to ask: how

409: large is the set of ``exceptional'' times at which percolation occurs?

410: In this direction, we have the following result.

411: 

412: \begin{theorem}\label{th.hd}

413: The Hausdorff dimension of the set of times 

414: at which dynamical critical site percolation

415: on the triangular lattice has an infinite cluster is 

416: an almost sure constant which lies in $[\frac{1}{6},\frac{31}{36}]$.

417: \end{theorem}

418: 

419: We conjecture that $\frac{31}{36}$ is the correct answer.

420: In a different direction, once we know that there are exceptional times at which 

421: percolation occurs, it is natural to ask how many clusters can exist at these

422: exceptional times. The following provides the answer. 

423: 

424: \begin{theorem} \label{th.no2clusters}

425: On the triangular lattice, a.s.\ there are no times at which 

426: dynamical critical site percolation has 2 or more infinite open clusters.

427: \end{theorem}

428: 

429: For the square grid, we can only prove

430: \begin{theorem} \label{th.no3clustersZ2}

431: On $\Z^2$, a.s.\ there are no times at which 

432: dynamical critical bond percolation has 3 or more infinite open clusters.

433: \end{theorem}

434: 

435: In some of the figures, we will represent open sites by white hexagons on the

436: dual grid, and

437: closed sites by black hexagons. Thus, percolation clusters

438: correspond to connected components of the union of the white hexagons.

439: These will also be called white clusters. Likewise, we may also consider

440: black clusters, which are connected components of black hexagons.

441: 

442: Asking whether 2 infinite white clusters can coexist

443: at some time is very different from asking whether 2 infinite clusters of 

444: {\sl different} colors can coexist at some time. We conjecture that there are 

445: in fact exceptional times 

446: at which there is both a white and a black infinite cluster and that the 

447: Hausdorff dimension of such times is $2/3$. We can however prove the following.

448: 

449: \begin{theorem} \label{th.differentclusters}

450: On the triangular lattice, a.s.\ the

451: Hausdorff dimension of the set of times at which there is both an infinite white

452: cluster and an infinite black

453: cluster is at most $2/3$.

454: \end{theorem}

455: 

456: We also have the following two results concerning the upper half plane.

457: 

458: \begin{theorem} \label{th.halfplane}

459: On the triangular lattice intersected with the upper half plane, a.s.\ 

460: the Hausdorff dimension of the set of times at which there is an

461: infinite cluster is at most $5/9$.

462: \end{theorem}

463: 

464: \begin{theorem} \label{th.halfplane2}

465: On the triangular lattice intersected with the upper half plane, a.s.\ the

466: set of times at which there is both an infinite white cluster 

467: and an infinite black cluster

468: is empty.

469: \end{theorem}

470: 

471: Theorems~\ref{th.no2clusters}, \ref{th.differentclusters},

472: \ref{th.halfplane} and \ref{th.halfplane2}

473: will follow immediately from generalizations presented

474: in the last part of the paper, which are concerned with studying dynamical percolation

475: on two other 2 dimensional objects, namely wedges and cones.

476: For every $\theta \in (0,\infty)$, we let $W_\theta$ denote the 

477: wedge of angle $\theta$ and $C_\theta$ denote the cone of angle $\theta$. 

478: For $C_\theta$, we will require that $\theta$ is a multiple of $\pi/3$.

479: The precise definitions of these will be given in Section \ref{sec:background}.

480: First, we mention that for all $\theta$, the critical value for site percolation

481: on $W_\theta$ and on $C_\theta$ is $1/2$, as for site percolation

482: on the triangular grid and bond percolation on $\Z^2$. 

483: 

484: The following results provide upper and lower bounds on the critical 

485: angle for which

486: there are exceptional times for certain $k$-arm type events as well as provide

487: estimates for the Hausdorff dimension of the set of exception times for a given

488: angle. In these results, if an upper bound on the Hausdorff dimension

489: is negative, this means that the set in question is empty. 

490: 

491: We will only do the case where the arms are alternating in color (and hence

492: for the case of cones, there will be one or an even number of arms).

493: We do this partially because

494: it is easier than the general case and because it is all that is needed

495: in order to make statements concerning the number of infinite clusters.

496: 

497: By a $k$-arm event, we mean an event of the form ``there are $k$ disjoint 

498: infinite paths having a specified color sequence''; 

499: for a wedge, the color sequence is well-defined while

500: for a cone, it is well-defined up to cyclic permutations.

501: 

502: 

503: \begin{theorem} \label{th:wedges}

504: Fix the wedge $W_\theta$ and for integer $k\ge 1$, let 

505: $A_{W_\theta}^k$ be the event that

506: there are $k$ infinite disjoint paths

507: in $W_\theta$ whose colors alternate. Then a.s.\ the Hausdorff 

508: dimension, $H_{W_\theta}^k$, of the set of exceptional times at which 

509: $A_{W_\theta}^k$ occurs satisfies

510: $$

511: 1-\frac{4k(k+1)\pi}{3\theta} \le  

512: H_{W_\theta}^k

513: \le

514: 1-\frac{2k(k+1)\pi}{9\theta}.

515: $$

516: In particular, for any $k\ge 1$, there are exceptional times

517: for the event $A_{W_\theta}^k$

518: for $\theta > \frac{4k(k+1)\pi}{3}$

519: and there are no exceptional times

520: for $\theta < \frac{2k(k+1)\pi}{9}$.

521: \end{theorem}

522: 

523: 

524: \begin{theorem} \label{th:cones}

525: Fix the cone $C_\theta$ with $\theta$ a multiple of $\pi/3$

526: and let, for $k= 1$ or $k>1$ even, 

527: $A_{C_\theta}^k$ be the event that

528: there are $k$ infinite disjoint paths in $C_\theta$

529: whose colors alternate (if $k>1$). 

530: Then a.s.\ the Hausdorff 

531: dimension, $H_{C_\theta}^k$, of the set of exceptional times at 

532: which $A_{C_\theta}^k$ occurs satisfies

533: $$

534: 1-\frac{5\pi}{3\theta} \le  

535: H_{C_\theta}^1

536: \le

537: 1-\frac{5\pi}{18\theta}

538: $$

539: and for $k\ge 2$

540: $$

541: 1-\frac{4(k^2-1)\pi}{3\theta} \le  

542: H_{C_\theta}^k

543: \le

544: 1-\frac{2(k^2-1)\pi}{9\theta}.

545: $$

546: In particular, for $k=1$, there are exceptional times

547: for the event $A_{C_\theta}^1$ for $\theta > \frac{5\pi}{3}$

548: and there are no exceptional times for $\theta < \frac{5\pi}{18}$,

549: while for $k\ge 2$, there are exceptional times

550: for the event $A_{C_\theta}^k$ for $\theta > \frac{4(k^2-1)\pi}{3}$

551: and there are no exceptional times for $\theta < \frac{2(k^2-1)\pi}{9}$.

552: \end{theorem}

553: 

554: Theorem \ref{th:cones} is presumably true for other values of $\theta$ 

555: provided that a proper definition of $C_\theta$ would be given.

556: 

557: \noindent

558: Remark:

559: One should note that the upper bounds on the Hausdorff

560: dimension given in Theorems~\ref{th.hd},\ref{th.differentclusters},\ref{th.halfplane},

561: \ref{th:wedges} and \ref{th:cones} are all of the form $1-(4/3)\xi$ where $\xi$ is the 

562: critical exponent for the given event.

563: 

564: \medskip

565: There is an abstract theory of L\'evy processes

566: on groups~\cite{Haw84,Eva89}, which gives a

567: criterion for a L\'evy process (such as $\omega_t$) to hit a set $A$

568: (such as the set of configurations which contain an infinite component). 

569: Basically, to show that $A$ is hit, one needs to prove that there exists a

570: probability measure $\mu$ on $A$ which has $\|\mu\|_*<\infty$

571: for an appropriate Hilbert norm $\|\cdot\|_*$, based on the Fourier

572: transform. It seems that we could use this framework in the present

573: paper, but that would not essentially simplify the core issues we deal with.

574: Moreover, it seems that our hands-on approach facilitates some generalizations,

575: which the L\'evy process theory does not cover, which brings us to

576: our next remark.

577: \medskip

578: 

579: The fact that the time between flips has an exponential

580: distribution is not really essential here,

581: and the results apply in greater generality.

582: Let $\omega_t(v)$ denote the indicator function for the

583: event that at time $t$ the site $v$ is white.

584:   Basically, all of the results 

585: concerning existence of exceptional times and lower bounds on Hausdorff 

586: dimension go through (with essentially the same proofs) in the more general setting 

587: where we assume that

588: 

589: \begin{enumerate}

590: \item[(i)] The processes $t\mapsto \omega_t(v)$ are independent (possibly with different

591: distributions depending on $v$) as $v$ runs over all sites. 

592: \item[(ii)] $\P[\omega_t(v)=1]=1/2$ for all $t$ and $v$. 

593: \item[(iii)] There is $c>0$ so that

594: $$

595: \bigl|\Eb{(-1)^{\omega_t(v)}(-1)^{\omega_s(v)}}\bigr| \le 1-c\, |t-s|

596: $$

597: for all $v$ and all $t$ and $s$ satisfying $|t-s|<c$.

598: \item[(iv)] For each $v$, the process $\omega_t(v)$  has

599: right continuous paths a.s. 

600: \end{enumerate}

601: (Condition (iv) is just a technical condition to insure that the events that we

602: consider are measurable.)

603: 

604: For results concerning upper bounds on Hausdorff dimension, the proofs go through assuming 

605: (i), (ii), (iv) and 

606: \begin{enumerate}

607: \item[(v)]  There is a $c>0$ so that

608: $$\E[\text{number of flips of }\omega_t(v) \text{ during }(t_1,t_2)] \le c (t_2-t_1)

609: $$

610: for all $v$ and all $t_1,t_2\in\R$ satisfying $t_1 < t_2<t_1+c$.

611: \end{enumerate}

612: However, for simplicity, we stick to the original setup.

613: 

614: \medskip

615: 

616: 

617: We mention a few other papers where analogous questions to those studied

618: in \cite{HPS} have been studied for other models. First, the results in

619: \cite{HPS} were extended and refined in \cite{PS}. Next, analogous questions

620: for the Boolean model where the points undergo independent Brownian motions was studied

621: in \cite{BMW}. Analogous questions for the lattice case for certain

622: interacting particle systems (where updates are not done in an independent fashion)

623: are studied in \cite{BrSt}. Finally in \cite{BeSc}, it is shown that

624: there are 

625: exceptional two dimensionl slices for the Boolean model in four dimensions.

626: 

627: The rest of the paper is organized as follows. 

628: In Section \ref{sec:dilute}, we will first provide background

629: on the Fourier-Walsh expansion of a function defined on $\{0,1\}^n$ as well

630: as connections with noise sensitivity and then continue on 

631: to give the proof of Theorem \ref{t.noise} as well as a 

632: generalization to the case where the algorithm is not required to always determine the value 

633: of the function $f$. (This will accomodate readers who are only interested in

634: Theorem \ref{t.noise}.) In Section \ref{sec:background}, we will give necessary background 

635: concerning percolation including a

636: discussion of critical site percolation on the triangular lattice as well as

637: a brief discussion of interfaces and critical exponents. 

638: In Section \ref{sec:percalg}, we will construct two 

639: algorithms determining certain events involving critical site percolation on the 

640: triangular lattice and analyze them to obtain upper bounds on the probability that 

641: a vertex is looked at during the algorithm. (For readers who only want to read

642: Theorem \ref{t.noise} and see how to apply it, they can just glance through 

643: Section \ref{sec:background} and then read Section \ref{sec:percalg}.)

644: Section~\ref{sec:percalg} also

645: gives a very detailed discussion of interfaces and

646: completes the proofs of Theorems \ref{th:crossingquantsquare} and 

647: \ref{th:crossingquant} by applying Theorem \ref{t.noise}. In 

648: Section \ref{sec:except}, we give the proof of Theorem \ref{th:except}, and

649: in Section \ref{sec:hd} we give the proof of Theorem \ref{th.hd}.

650: (Although the upper bound of $31/36$ given in Theorem \ref{th.hd} is a 

651: special case of Theorem \ref{th:cones}, we choose to give a

652: different direct proof of this 

653: without reference to the work done in Section \ref{sec:wedgesupper}.)

654: In Section \ref{sec:wedgeslower}, we prove the lower bounds 

655: on the Hausdorff dimension stated in

656: Theorems \ref{th:wedges} and \ref{th:cones}. 

657: In Section \ref{sec:wedgesupper}, we prove the upper 

658: bounds on the Hausdorff dimension

659: stated in Theorems \ref{th:wedges} and \ref{th:cones}. 

660: This will be based on a 

661: general formula which gives an upper bound on the 

662: Hausdorff dimension of various random sets (or proves they are empty)

663: in terms of {\sl influences} (Theorem~\ref{th.hdgeneral}).

664: We conclude the section by showing that

665:  Theorems \ref{th.no2clusters},

666: \ref{th.differentclusters}, \ref{th.halfplane} and \ref{th.halfplane2} 

667: immediately follow from Theorems \ref{th:wedges} and \ref{th:cones}.

668: After this, in Section \ref{sec:square},

669: we prove Theorem~\ref{th.no3clustersZ2}

670: and explain several plausible

671: ways in which the proof of Theorem \ref{th:except}

672: might be extended to bond percolation on the square lattice. 

673: In Section \ref{sec:questions}, we present some open questions.

674: 

675: Finally, the appendix proves some 

676: results about (non-dynamical) critical percolation

677: that are needed for Theorems~\ref{th.no2clusters}--\ref{th:cones}.

678: The main result is that if $r<r'<r''$, then the probability to have

679: $j$ crossings in a prescribed color sequence between distances $r$ and $r''$

680: from $0$ is equal to the product of the corresponding probabilities

681: between radii $r$ and $r'$ and between radii $r'$ and $r''$, times an error

682: term (depending on $j$) that is bounded away from $0$ and infinity.

683: Another consequence is that one gets good control on the positions

684: of the crossings at the inner and outer radii, as was already

685: demonstrated by Kesten~\cite[Lemma 2]{K2}. The proofs in the appendix

686: also establish the corresponding statements for critical bond percolation

687: in $\Z^2$.

688: 

689: 

690: 

691: \section{Noise sensitivity of algorithmically dilute functions}\label{sec:dilute}

692: 

693: 

694: In this section, we give some background and then prove Theorem \ref{t.noise}.

695: 

696: \subsection{Noise sensitivity background}\label{ss.noisebackground}

697: 

698: For a function $f$ from $\Omega=\Omega_n:=\{0,1\}^n$ to $\R$,

699: the Fourier-Walsh expansion of $f$ is given by

700: $ f = \sum_{S \subseteq [n]} \hat{f}(S) \chi_S,$ 

701: where, $\chi_S(T)=(-1)^{|S \cap T|}$ and $\hat{f}(S)= \int f \chi_S$.

702: Here and in the following, $\int$ refers to integration with respect to

703: uniform measure and we identify any vector 

704: $x\in\Omega_n$ with the subset $\{j\in[n]:x_j=1\}$ of

705: $[n]=\{1,2,\dots,n\}$.

706: Consequently, $|x|$ denotes the cardinality of that set; that is,

707: $|x|=\|x\|_1$ for $x\in\Omega_n$. The $\{\chi_S\}_{S\subseteq [n]}$ are an 

708: orthornormal basis for the $2^n$ dimensional vector space of functions from

709: $\Omega_n$ to $\R$. In particular,

710: $$

711: \|f\|^2=\sum_{S \subseteq [n]} \hat{f}(S)^2.

712: $$

713: 

714: We now generalize the definition of $N(A,\eps)$ given in the introduction

715: to any function $f:\Omega\to\R$ by defining

716: $$

717: N(f,\eps):=

718: \var\Bigl[\Eb{f(Y_1,\ldots,Y_{m})|X_1,\ldots,X_{m}}\Bigr].

719: $$

720: It is easy to see that (see page 14 in \cite{BKS}) 

721: \begin{equation}\label{e.page14}

722: N(f,\eps)= \sum_{\emptyset\neq S \subseteq [n]} \hat{f}(S)^2 (1-2\epsilon)^{2|S|}.

723: \end{equation}

724: This explains the importance of the Fourier-Walsh expansion in the study of

725: noise sensitivity.

726: 

727: \subsection{Proof of Theorem \ref{t.noise}}\label{ss.noisetheorem}

728: 

729: Before giving the proof, we discuss some heuristics.

730: One may first believe that an estimate such as~\eref{e.noise} would

731: be valid because when the algorithm terminates, the value of

732: $f$ is completely determined, and hence perhaps all the nonzero

733: Fourier coefficients $\hat f(S)\ne 0$, $S\ne\emptyset$, must

734: satisfy $S\cap \II\ne\emptyset$. However, this is easily shown not 

735: to be the case. At the $t$-th step of the algorithm,

736: after $t$ bits have been determined,

737: we may consider a new function $f_t$, which is $f$ 

738: with those determined bits substituted. 

739: If at the $(t+1)$-th step, the $i$-th bit $\omega_i$

740: of the input $\omega\in\Omega$ is examined, then

741: in the passage from $f_t$ to $f_{t+1}$, there is a collapsing

742: of Fourier coefficients: 

743: $\hat f_{t+1}(S)=\hat f_t(S)+(-1)^{\omega_i} \hat f_t(S\cup\{i\})$

744: and $\hat f_{t+1}(S\cup\{i\})=0$

745: for every $S\subseteq\nn\setminus\{i\}$.

746: Thus, the coefficient $\hat f_{t+1}(S)$ may vanish 

747: when some bit $i\in S$ is examined by time $t+1$ or when

748: some $i\notin S$ is chosen at time $t+1$ and it happens that

749: $\hat f_t(S)+(-1)^{\omega_i} \hat f_t(S\cup\{i\})=0$.

750: The latter, which we call \lq\lq collapsing from

751: above\rq\rq, may seem like a highly nongeneric 

752: situation. However, we cannot rule it out because

753: we are primarily interested in very non-generic functions,

754: namely, functions with values in $\{0,1\}$.

755: The proof below uses a simple decomposition argument to

756: handle the possibility of

757: collapsing from above.

758: 

759: In the following, we let $\tilde{\Omega}$ denote the probability

760: space that includes the randomness in the input bits of $f$ and the randomness

761: used to run the algorithm and we let $\E$ denote the

762: corresponding expectation.

763: Without loss of generality, elements of 

764: $\tilde{\Omega}$ can be represented as

765: $\tilde{\omega}=(\omega,\tau)$ where 

766: $\omega$ are the random bits and $\tau$ represents the

767: randomness necessary to run the algorithm.

768: 

769: \proof

770: Fix $k\ge 1$. Let 

771: $$

772: g(\omega):=\sum_{|S|=k} \hat f(S)\,\chi_S(\omega)\,,\qquad

773: \omega\in\Omega.

774: $$

775: The left hand side of~\eref{e.noise} is equal to $\|g\|^2$.

776: Let $\II\subset\nn$ be the random set of all bits examined by the algorithm.

777: Let $\ev A$ denote the 

778: minimal $\sigma$-field for which $\II$ is measurable and every

779: $\omega_i$, $i\in \II$, is measurable; this can be viewed as the

780: relevant information gathered by the algorithm.

781: For any function $h:\Omega\to\R$, let

782: $h_\II:\Omega\to\R$ denote the random function obtained by substituting

783: the values of the bits in $\II$. 

784: More precisely, if $\tilde{\omega}=(\omega,\tau)$

785: and $\omega'\in\Omega$, then $h_\II(\tilde{\omega})(\omega')$ is

786: $h(\omega'')$ where $\omega''$ is $\omega$ on $\II(\tilde{\omega})$ and is

787: $\omega'$ on $[n]\backslash \II(\tilde{\omega})$.

788: In this way, $h_\II$ is a random variable

789: on $\tilde{\Omega}$ taking values in the set of mappings from

790: $\Omega$ to $\R$ and it is immediate that this random variable is

791: $\ev A$-measurable.

792: When the algorithm terminates, the unexamined bits in $\Omega$

793: are unbiased and hence 

794: $\Eb{h\md \ev A}=\int h_\II(=\hat {h_\II}(\emptyset))$

795: where $\int$ is defined, as usual, to be integration with

796: respect to uniform measure on $\Omega$.

797: It follows that $\Es{h}=\Es{\int h_\II}$.

798: 

799: More generally, if $u:\R\to\R$, then

800: $(u\circ h)_\II=u\circ h_\II$ 

801: and hence, as above, $ \Es{u(h)}=\EB{\int u(h_\II)}$. 

802: In particular, for all $h$,

803: \begin{equation}\label{e.equiv}

804: \|h\|^2=\Eb{h^2}=\EB{\int h_\II^2}=\Eb{\|h_\II\|^2}.

805: \end{equation}

806: 

807: Since the algorithm determines  $f$, it is $\ev A$ measurable,

808: and we have

809: \begin{equation*}

810: \|g\|^2=\Es{g\,f}=\EB{\Eb{g\,f\md \ev A}}=

811: \EB{f\,\Eb{g\md\ev A}}.

812: \end{equation*}

813: Since $\Eb{g\md \ev A}=\hat {g_\II}(\emptyset)$,  Cauchy-Schwarz therefore gives

814: \begin{equation}\label{e.ept}

815: \|g\|^2 \le \sqrt{ \Es{\hat g_\II(\emptyset)^2}}\,\|f\|\,.

816: \end{equation}

817: We may write,

818: $$

819: \Es{\hat g_\II(\emptyset)^2}

820: =

821: \Eb{\|g_\II\|^2} -

822: \EB{\sum_{|S|>0} \hat g_\II(S)^2}.

823: $$

824: This and~\eref{e.equiv} with $h=g$ imply that

825: \begin{multline*}

826: \Es{\hat g_\II(\emptyset)^2}

827: \le

828: \|g\|^2

829: -\EB{\sum_{|S|=k}\hat g_\II(S)^2}

830:  \\

831: =

832: \sum_{S\subseteq\nn}\hat g(S)^2

833: -\EB{\sum_{|S|=k}\hat g_\II(S)^2}

834: =

835: \sum_{|S|=k}\Eb{\hat g(S)^2-\hat g_\II(S)^2}

836: \,.

837: \end{multline*}

838: It is easily seen that for any function $h$,

839: $h_\II=\sum_{S}\hat h(S)\,(\chi_S)_\II$.

840: We apply this with $h=g$.

841: Since $\hat g(S')=0$ if $|S'|> k$,

842: it follows that for all

843: $S\subset\nn$ satisfying $|S|=k$

844: $$

845: \hat g_\II(S)=

846: \begin{cases}

847: \hat g(S),& S\cap \II=\emptyset, \\

848: 0, &S\cap \II\ne\emptyset\,.

849: \end{cases}

850: $$

851: The above estimate for

852: $\Es{\hat g_\II(\emptyset)^2}$ therefore gives

853: $$%\begin{multline*}

854: \Es{\hat g_\II(\emptyset)^2}

855: \le

856: \sum_{|S|=k}\hat g(S)^2\,\Pb{S\cap \II\ne\emptyset}

857: \le \sum_{|S|=k}\hat g(S)^2\,\sum_{i\in S}\Ps{i\in \II}

858: \le \|g\|^2\,k\,\delta

859: \,.

860: $$%\end{multline*}

861: Substituting this estimate in~\eref{e.ept} and squaring the resulting

862: inequality completes the proof of the theorem.

863: \QED

864: 

865: 

866: The theorem may be easily generalized to situations where the

867: algorithm does not always determine the value of $f$ precisely;

868: that is, $f_\II(x)$ still depends on $x\in\Omega$.

869: 

870: Set 

871: $$

872: \var_\Omega(f_\II):=\int (f_\II)^2 - \Bigl(\int f_\II\Bigr)^2

873: =\sum_{S\ne\emptyset}\hat f_\II(S)^2\,,

874: $$

875: where the integrations are with respect to the uniform

876: probability measure on $\Omega$.

877: Note that $\Eb{\var_\Omega(f_\II)}$ is an indicator for how precisely

878: the algorithm can be used to approximate $f$;

879: when $\var_\Omega(f_\II)$ is small, with high conditional probability,

880: $|\hat f_\II(\emptyset)-f|$ is not too large.

881: 

882: When $\var_\Omega(f_\II)\ne0$, we have to replace the calculation

883: in the proof of Theorem~\ref{t.noise}

884: by the following

885: \begin{multline}

886: \|g\|^2=

887: \Eb{g_\II\,f_\II}

888: =\Eb{ g_\II\,\hat f_\II(\emptyset)}

889: + 

890: \Eb{g_\II\,(f_\II-\hat f_\II(\emptyset))}

891: \\

892: \le

893: \Bigl(\Es{\hat g_\II(\emptyset)^2}\,\Es{\hat f_\II(\emptyset)^2}\Bigr)^{1/2}

894: +

895: \Bigl(\Es{ g_\II^2}\,\Es{\var_\Omega( f_\II)}\Bigr)^{1/2}. 

896: \end{multline}

897: Since $\Eb{\hat f_\II(\emptyset)^2+\var_\Omega(f_\II)}=\Es{f^2}$,

898: we have $\Es{\hat f_\II(\emptyset)^2}\le \|f\|^2$.

899: Thus, the above gives

900: $$

901: \|g\|^2 \le

902: \sqrt{\Eb{\hat g_\II(\emptyset)^2}}\,\|f\|

903: +

904: \|g\|\,\sqrt{\Es{\var_\Omega(f_\II)}}\,.

905: $$

906: Using the same estimate for $\Eb{\hat g_\II(\emptyset)^2}$ as in the proof

907: of Theorem~\ref{t.noise}, we  obtain

908: $$

909: \|g\|^2

910: \le

911: \|g\|\,\|f\|\sqrt{k\,\delta}+

912: \|g\|\,\sqrt{\Es{\var_\Omega(f_\II)}}\,.

913: $$

914: Consequently, squaring both sides gives the following

915: generalization of~\eref{e.noise}:

916: \begin{multline}\label{e.gnoi}

917: \sum_{|S|=k}\hat f(S)^2

918: \le

919: \Bigl(\|f\|\sqrt{k\,\delta}+

920: \sqrt{\Es{\var_\Omega(f_\II)}}\Bigr)^2

921: \\

922: \le

923: 2\,k\,\delta\,\|f\|^2+

924: 2\,

925: \Es{\var_\Omega(f_\II)}\, .

926: \end{multline}

927: \bigskip

928: 

929: Theorem~\ref{t.noise} holds more generally than stated.

930: If $W$ is a random subset of $\nn$,

931: we say that $W$ is a {\bf witness} for $f:\Omega\to\R$ if

932: the value of $f$ is determined by its restriction to

933: $W$.  We say that $W$ is {\bf $\delta$-dilute} if

934: $\max_{i\in\nn}\Pb{i\in W(\omega)}\le\delta$.

935: The related notion of short witnesses is of central importance

936: in Talagrand's epic isoperimetric saga~\cite{Talagrand}.

937: The proof of Theorem~\ref{t.noise} holds in the more general setting where the

938: random

939: set $J$ is a witness

940: with the property that for all subsets $A\subseteq \nn$,

941: conditioned on $J=A$ (assuming this has positive probability)

942: and conditioned on $\omega$ restricted to $A$, the

943: $\{\omega_i\}_{i\not \in A}$ are uniform i.i.d.\ bits.

944: As pointed out to us by Asaf Nachmias, 

945: Theorem~\ref{t.noise} does not hold for arbitrary witnesses,

946: even if we allow for a multiplicative constant in the

947: right hand side of~\eref{e.noise}: if you take \lq\lq Recursive Ternary Majority\rq\rq\ on

948: $n=3^h$ bits, there is a (symmetric) witness having only $2^h$ elements,

949: yielding a $\delta$ which is $(2/3)^h$; however, the sum of the squares

950: of the level 1 Fourier coefficients is $(3/4)^h$.

951: 

952: \section{Percolation background and notations}\label{sec:background}

953: 

954: Duality plays a central role in the theory of percolation in two dimensions.

955: A dual-open path on the triangular grid is defined as a path

956: in the grid whose vertices are all closed. For the square grid,

957: a dual-open path is defined as a path in the dual of the square grid that

958: does not intersect any open edge in the primal grid.

959: The basic observation is that for site percolation on the 

960: triangular grid at $p=1/2$ the distribution of the collection of

961: dual-open paths is the same as the distribution of the collection of primal open paths.

962: (Sometimes, we use the term ``primal open path'', for an open path, to make

963: the distinction with the dual-open path clearer.)

964: For critical bond percolation on the square grid at $p=1/2$, the distribution

965: of the dual-open paths is the image of the distribution of the

966: open paths under translation by $(1/2,1/2)$.

967: This simple duality is one of the important ingredients in the proof by

968: Kesten that $p_c=1/2$ for these two percolation models~\cite[pg.~53]{Kesten}

969: and the earlier proof by Harris (see \cite{Harris})

970: that there is a.s.\ no infinite cluster at $p=1/2$.

971: 

972: For $0\le r <R<\infty$, let $A(r,R)$ denote the event that there is an open

973: crossing of the annulus $r\le|z|\le R$, namely, an open path connecting a vertex

974: inside the disk $|z|\le r$ to a vertex in $|z|\ge R$.

975: Let $\alpha(r,R)$ denote the probability of

976: ${A(r,R)}$ at percolation parameter $p=p_c=1/2$. 

977: Abbreviate $\alpha(0,R)$ by $\alpha(R)$.

978: For convenience, we adopt the convention $\alpha(r,R)=1$ whenever $r\ge R$.

979: The function $\alpha(r,R)$ is essentially multiplicative, in the following

980: sense: there is a constant $C>0$ such that for every $0\le r_1\le r_2\le r_3<\infty$,

981: \begin{equation}\label{e.multip}

982: C^{-1}\,\alpha(r_1,r_3)\le \alpha(r_1,r_2)\,\alpha(r_2,r_3)\le C\,\alpha(r_1,r_3)\,.

983: \end{equation}

984: In fact, this holds for critical bond percolation on the square grid as well

985: as for critical site percolation on the triangular grid.

986: The (standard) proof of~\eref{e.multip}

987: is based on the Harris-FKG inequality

988: %(see \cite{Grimmett}, page 34)

989: and the celebrated Russo-Seymour-Welsh (RSW) theorem

990: (see~\cite{Grimmett,Kesten}).

991: %(see \cite{Grimmett}, page 315).

992: A proof of a generalization of~\eref{e.multip} is given in the appendix.

993: Another consequence of RSW that we will use is the existence of a constant

994: $c>0$ such that for every $r>0$,

995: \begin{equation}\label{e.rsw}

996: c\le \alpha(r,2\,r)\,.

997: \end{equation}

998: 

999: The Stochastic L\"owner evolution (SLE) introduced in~\cite{Schramm} is a one parameter 

1000: family of random curves indexed by a real positive parameter $\kappa$.

1001: It was conjectured in \cite{Schramm} 

1002: that the scaling limit of outer boundaries of critical

1003: site percolation clusters on the triangular grid

1004: as well as

1005: bond percolation clusters on $\Z^2$ 

1006: are (chordal) SLE$_\kappa$ with ${\kappa=6}$.

1007: Smirnov~\cite{Smirnov,Smirnov1} proved the corresponding statement for critical site

1008: percolation on the triangular lattice. 

1009: (See also \cite{CN}.) We now explain some of this more precisely. We first

1010: perform independent site percolation on the upper half of the triangular lattice 

1011: but declare the sites $\{(k,0):k > 0\}$ to all be open and $\{(k,0):k\le 0\}$ to all be closed.

1012: In the hexagonal grid dual to the triangular grid, there will then be a

1013: unique path in the upper half plane from $(\frac{1}{2},0)$ to $\infty$ 

1014: which has white hexagons containing open 

1015: sites on the right and black hexagons containing closed

1016: sites on the left. (See Figure~\ref{f.pcurve}.)

1017: Smirnov's result is that the limit (in an appropriate toplogy) as the mesh size of the

1018: lattice goes to $0$ of the law of this path

1019: is chordal SLE$_6$.

1020: This path described above, which has open sites

1021: on its right and closed sites on its left, is an example of what is called an 

1022: {\it interface}.

1023: 

1024: \begin{figure}

1025: \centerline{\epsfysize=2.5in\epsfbox{pcurve.ps}}

1026: \begin{caption} {\label{f.pcurve}The percolation interface.}

1027: \end{caption}

1028: \end{figure}

1029: 

1030: The conformal invariance and the SLE description of critical percolation on 

1031: the triangular lattice allowed researchers to prove a number of conjectures by 

1032: physicists concerning so-called {\sl critical exponents}.

1033: For example, for critical site percolation on the triangular grid, it was established in 

1034: \cite{LSW} that 

1035: \begin{equation}\label{e.onearm}

1036: \alpha(R)=R^{-5/48 + o(1)} \text{ as $R\to\infty$}\,.

1037: \end{equation}

1038: In fact, the same proof actually gives for $R\ge r\ge 1$,

1039: \begin{equation}\label{e.onearmstrong}

1040: \alpha(r,R)=(R/r)^{-5/48 + o(1)} \text{ as $R/r\to\infty$}\,.

1041: \end{equation}

1042: % By a slight abuse of notations, here and in the following,

1043: % an expression of the form $s^{o(1)}$

1044: % will stand for any quantity $f$ such that there

1045: % is a function $g(s)$ of $s$ only satisfying

1046: % $g(s)^{-1}\le f\le g(s)$ and $\lim_{s\to L} \log g(s)/\log s=0$,

1047: % where $L=0+$ or $L=\infty$, depending on the context.

1048: 

1049: For $1\le r \le R$, the two arm function $\alpha_2(r,R)$ denotes the probability

1050: that there is both an open path from $|z|\le r$ to $|z|\ge R$ and also a 

1051: dual-open path from $|z|\le r$ to $|z|\ge R$. 

1052: We abbreviate $\alpha_2(1,R)$ by $\alpha_2(R)$.

1053: 

1054: Next, let $M$ be a half plane in $\R^2$ and let $v$ be some vertex in $M$

1055: satisfying $\dist (v,\p M)\le 2$.

1056: Denote by $\alpha^+(R)$ the probability that there is an open

1057: path in $M$ from $v$ to distance at least $R$ away from $v$.

1058: This quantity depends on $R$, $v$, and $M$, but the dependence on

1059: $v$ and $M$ will usually be suppressed.

1060: More generally, for $0\le r< R$, let 

1061: $\alpha^+(r,R)$ denote the probability that there is

1062: an open path in $M$ from some vertex $u$ satisfying $|u-v|\le r$ to some 

1063: vertex $w$ satisfying $|w-v|\ge R$.

1064: 

1065: As with $\alpha(r,R)$, we adopt the convention

1066: $\alpha^+(r,R)=1$ whenever $r\ge R$.

1067: 

1068: It is known that

1069: the functions $\alpha^+$ and $\alpha_2$ also satisfy~\eref{e.multip}, with possibly

1070: different constants. (When considering $\alpha^+$,

1071: this applies to any fixed choice of $M$ and $v$.)

1072: These inequalities are valid for site percolation on the triangular grid as well

1073: as bond percolation on the square grid. 

1074: A proof can be found in the appendix.

1075: For site percolation on the triangular grid, the corresponding exponents

1076: were established in~\cite{SW}:

1077: \begin{equation}\label{e.expo}

1078: \alpha^+(R)=R^{-1/3+o(1)},\qquad

1079: \alpha_2(R)=R^{-1/4+o(1)},

1080: \end{equation}

1081: as $R\to\infty$.

1082: In fact, the same proofs actually give for $R\ge r\ge 1$

1083: \begin{equation}\label{e.expostrong}

1084: \alpha^+(r,R)=(R/r)^{-1/3+o(1)},\qquad

1085: \alpha_2(r,R)=(R/r)^{-1/4+o(1)},

1086: \end{equation}

1087: as $R/r\to\infty$.

1088: 

1089: For bond percolation on the square grid, such exact estimates are unavailable,

1090: because there is currently no proof that the interface converges

1091: to SLE$_6$.

1092: In the case of the square grid, the

1093: estimate $\alpha(r,R)\le C\,(R/r)^{-\eps}$,

1094: where $C,\eps>0$ are constants, follows easily from the 

1095: RSW theorem (see \cite{Grimmett,Kesten}).

1096: The RSW proofs can give an actual value for $\eps$, but it is rather

1097: small. We can also use the obvious estimate $\alpha^+(r,R)\le\alpha(r,R)$

1098: to obtain a similar bound for $\alpha^+$.

1099: 

1100: \smallskip

1101: 

1102: We now give the precise definitions for wedges and cones.

1103: For this purpose, we first recall the definition of the infinitely

1104: branched cover of $\R^0$ over $0$.

1105: Let $X=\bl\{(z,\theta):z\in\C\setminus \{0\},\,\theta\in\R,e^{i\theta}|z|=z\br\}$,

1106: and set $\psi(z,\theta)=z$. On the surface $X$ we define the

1107: metric $d_X$ as the pullback of the Euclidean metric of $\R^2$ under

1108: $\psi$, namely, $d_X(x,y)$ is the infimum of the length

1109: of $\psi\circ \gamma$ for any continuous path $\gamma\subset X$ connecting $x$

1110: and $y$. Let $C_\infty$ denote the completion of $(X,d_X)$.

1111: Since $\R^2$ is complete, it is easy to see that

1112: $C_\infty\setminus X$ consists of a single point, which we denote by $0$.

1113: We extend the map $\psi$ by setting $\psi(0)=0$.

1114: Let $V$ be the set of points in $C_\infty$ that are mapped to vertices

1115: of the triangular grid under $\psi$. The triangular grid

1116: on $C_\infty$ has vertices $V$ and an edge between any two vertices

1117: at distance $1$ apart.

1118: Now the wedge $W_\theta\subset C_\infty$ is defined by

1119: $W_\theta:=\{0\}\cup \bl\{(z,\theta')\in X: \theta'\in[0,\theta)\br\}$.

1120: The triangular grid on $W_\theta$ is just the intersection of

1121: the triangular grid on $C_\infty$ with $W_\theta$.

1122: 

1123: On $C_\infty$ we may define the rotation $R_\theta$

1124: by $R_\theta(0)=0$ and $R_\theta(z,\theta')=(e^{i\theta}z,\theta+\theta')$.

1125: This is clearly an isometry of $C_\infty$.

1126: The cone $C_\theta$ is defined as the quotient $C_\infty/R_\theta$; that is,

1127: the set of equivalence classes of points in $C_\infty$, where two points

1128: are considered equivalent if one is mapped to the other by a power of

1129: $R_\theta$.  Now suppose that $\theta=n\,\pi/3$ where $n\in\N_+$. Then

1130: $R_\theta$ restricts to an isomorphism of the triangular grid on $C_\infty$.

1131: In this case we define the triangular grid on $C_\theta$ as the quotient

1132: of the grid on $C_\infty$ under $R_\theta$.

1133: In other words, the vertices are equivalence classes of vertices in

1134: $C_\infty$ and an edge appears between two equivalence classes if

1135: there is an edge connecting representatives of these classes.

1136: Note that $C_{2\pi}$ is just the euclidean plane with the triangular

1137: lattice.

1138: 

1139: \smallskip

1140: 

1141: 

1142: 

1143: We end this section with describing the so-called full and half plane 

1144: exponents for $k$-arm events that were derived in \cite{SW}.

1145: 

1146: For integer $k\ge 1$, let $A^k(r,R)$ be the event that there are $k$ disjoint

1147: crossings of the annulus $\{z\in\R^2: r\le |x|\le R\}$ with a specified color

1148: sequence (up to rotations), where we require that both colors appear 

1149: in the color sequence.

1150: For $k\ge 2$, and $r\ge 10k$, it was proved in \cite{SW} that

1151: \begin{equation}\label{e.fullpk}

1152: \alpha_k(r,R):=\Pb{A^k(r,R)}=(R/r)^{\frac{1-k^2}{12} + o(1)},

1153: \end{equation}

1154: as $R\to\infty$ while $r$ is fixed.

1155: (The result for $\alpha_2(R)$ in~\eref{e.expo} above is a special case of this.)

1156: Next, letting $A_+^k(r,R)$ be the event that there are $k$ disjoint paths in the 

1157: upper half plane from $|z|\le r$ to $|z|\ge R$ with any specified color sequence, then for

1158: $k\ge 1$, and $r\ge 10k$, it was proved in \cite{SW} that

1159: \begin{equation}\label{e.halfpk}

1160: \alpha^+_k(r,R):=\Pb{A_+^k(r,R)}=(R/r)^{\frac{-k(k+1)}{6} + o(1)},

1161: \end{equation}

1162: as $R\to\infty$ while $r$ is fixed. %

1163: (The result for $\alpha_+(R)$ in~\eref{e.expo} is a special case of this.) 

1164: Just as we said that the proofs of~\eref{e.expo} actually yield~\eref{e.expostrong},

1165: it is also the case that the proofs of~\ref{e.fullpk} and~\ref{e.halfpk} also

1166: yield versions when $R/r\to\infty$ while $r\ge 10\,k$ is not necessarily fixed.

1167: 

1168: 

1169: \section{Noise sensitivity for percolation} \label{sec:percalg}

1170: 

1171: \subsection{Simply connected case}\label{ss.simply}

1172: 

1173: To apply Theorem~\ref{t.noise} to percolation, we will need to describe algorithms achieving

1174: small revealment. One result of that nature is

1175: 

1176: 

1177: \begin{theorem}\label{t.square}

1178: Let $\cro=\cro_R$ be the indicator function for the 

1179: event that critical site percolation on the standard triangular grid

1180: contains a left to right crossing in some grid-approximating domain $D$

1181: to a large square of side length $R$.

1182: (For example, we could take $D$ to be the union

1183: of the hexagons in the dual grid that are contained in the

1184: square.)

1185: Then there is a randomized algorithm $A$ determining $\cro$ 

1186: such that $\delta_A \le R^{-1/4+o(1)}$

1187: as $R\to\infty$.

1188: 

1189: For critical bond percolation on the square grid, there is such an

1190: algorithm satisfying $\delta\le C\,R^{-a}$ for some constants

1191: $a,C>0$.

1192: \end{theorem}

1193: 

1194: \noindent

1195: Remark:

1196: Theorem~\ref{t.square} says that there is an algorithm for the

1197: relevant event which exposes on average at most $R^{7/4+o(1)}$ bits.

1198: Since the probability of points not too close to the boundary

1199: being pivotal is about $R^{-5/4+o(1)}$ (this is the 4 arm event)

1200: and for a monotone function $f$, $\hat f(\{i\})$ is

1201: the probability that $x_i$ is pivotal,

1202: the case $k=1$ in Theorem~\ref{t.noise} implies that the revealment

1203: is at least $R^{-1/2+o(1)}$.

1204: As pointed out in Peres, Schramm, Sheffield and Wilson~\cite{PSSW}, 

1205: this can also be obtained using an inequality of O'Donnell and Servedio.

1206: 

1207: Theorems~\ref{t.noise} and \ref{t.square} immediately give

1208: 

1209: \begin{corollary}\label{c.square}

1210: For every $\eps>0$ there is a constant $C=C(\eps)$ such that

1211: $$

1212: \sum_{|S|=k}\hat\cro_R(S)^2 \le C\, k\,R^{-1/4+\eps}

1213: $$

1214: holds for every $k=1,2,\dots$ and for every $R>0$.

1215: \QED

1216: \end{corollary}

1217: 

1218: The basic idea of the proof of Theorem~\ref{t.square} is rather simple.

1219: First we consider the interface started at the lower right corner and

1220: stopped when it hits the upper or left edges. (See Figure~\ref{f.interface}.)

1221: This interface is sufficient to determine $Q$.

1222: If we traverse the interface, revealing just the bits necessary for its

1223: determination, then with high probability most bits will not be

1224: examined. However, this does not yield an algorithm with small revealment

1225: because the hexagons near the lower right corner are very likely

1226: to be examined. 

1227: To rectify this problem, we instead start the interface at a different

1228: (random) location $p_0$ on the right side

1229: of $D$. 

1230: This determines the existence of a crossing from the right side above

1231: $p_0$ to the left side. Then another interface started at $p_0$ will

1232: determine if there is a crossing that starts below $p_0$.

1233: 

1234: \begin{figure}

1235: \centerline{\epsfysize=2in\epsfbox{sqpcurve.eps}}

1236: \begin{caption} {\label{f.interface}Following the interface from the corner.}

1237: \end{caption}

1238: \end{figure}

1239: 

1240: Let us now be a bit more precise regarding the notion of an interface.

1241: In the following, we use an equivalent dual version of the

1242: site percolation model on the triangular grid. The dual graph is

1243: the hexagonal grid, and we color the hexagon white if the site

1244: contained in it is open and black if the site in it is closed.

1245: Of course, there is no essential difference between these two

1246: representations. The advantage of this dual framework is that the

1247: figures are clearer and the notion of the interface is slightly more

1248: natural.

1249: 

1250: Note that bond percolation on the square grid also has a similar

1251: coloring representation. One such scheme is to

1252: color the squares of sidelength $1/2$ centered at the sites

1253: of $\Z^2$ white, color the squares of sidelength $1/2$ that are

1254: concentric with the square faces of $\Z^2$ black,

1255: and color each square of sidelength $1/2$ whose center is the

1256: midpoint of an edge of $\Z^2$ white or black if that edge is

1257: open or closed, respectively. See Figure~\ref{f.squareColorPerc}.

1258: This scheme has the important property that the boundary between

1259: the union of the white clusters is a $1$-manifold; that is,

1260: at every vertex of the grid $(1/2)\,\Z^2+(1/4,1/4)$ there are

1261: two edges that are on the common boundary.

1262: 

1263: \begin{figure}

1264: \centerline{\epsfysize=2.5in\epsfbox{sqPerc.eps}}

1265: \begin{caption} {\label{f.squareColorPerc}A color scheme for bond percolation on $\Z^2$.}

1266: \end{caption}

1267: \end{figure}

1268: 

1269: We now assume that $D$ is a bounded simply connected domain which is

1270: the interior of a union of hexagons in the hexagonal grid.

1271: Suppose that $p_0$ is a point on $\p D$ that is on the boundary

1272: of a single hexagon in $D$, and that $\zeta$ is a closed arc

1273: in $\p D\setminus\{p_0\}$.

1274: The interface in $\closure D$ from $p_0$ to $\zeta$ is a 

1275: random path $\beta$ contained in the $1$-skeleton of the hexagonal grid

1276: starting at $p_0$ and ending at a point in

1277: $\zeta$, as indicated in Figure~\ref{f.interfaceDef}.

1278: We can precisely define $\beta$ as the unique oriented simple path

1279: from $p_0$ to $\zeta$ that is contained in

1280: the union of the boundaries of the hexagons

1281: contained in $\closure D$ and satisfies

1282: (1) $\beta\cap\zeta$ consists of

1283: the terminal point of $\beta$,

1284: (2) whenever $\beta$ traverses an arc along the boundary

1285: of a black hexagon $H\subset\closure D$, the 

1286: arc is traversed counterclockwise around $\p H$, and

1287: (3) whenever $\beta$ traverses an arc along the boundary

1288: of a white hexagon $H\subset\closure D$, the 

1289: arc is traversed clockwise around $\p H$.

1290: It is easy to verify that this uniquely defines $\beta$,

1291: as follows.

1292: First, the initial arc of $\beta$ is determined by the

1293: color of the hexagon in $\closure D$ containing $p_0$.

1294: When $\beta$ first meets a hexagon contained in 

1295: $\closure D$, its turn is clearly specified.

1296: (If the hexagon is black, then $\beta$ must make

1297: a $\pi/3$ turn to the right,

1298: and if the hexagon is white, then $\beta$

1299: must make a $\pi/3$ turn to the left.)

1300: Now consider the situation in which $\beta$

1301: first meets a hexagon that is not contained in

1302: $\closure D$. Let $\beta'$ be the

1303: arc of $\beta$ from $p_0$ up to that point.

1304: Then $\beta$ at that point makes the turn into

1305: the component of $\closure D\setminus \beta'$

1306: that contains $\zeta$, as in the figure.

1307: 

1308: \begin{figure}

1309: \SetLabels

1310: \B(.56*.98)$\zeta$\\

1311: \R(.19*.48)$p_0$\\

1312: \endSetLabels

1313: %\ShowGrid

1314: \centerline{\epsfysize=2in%

1315: \AffixLabels{%

1316: \epsfbox{interfaceDef.eps}}%

1317: }

1318: \begin{caption} {\label{f.interfaceDef}An interface started at $p_0$

1319: and headed towards $\zeta$.}

1320: \end{caption}

1321: \end{figure}

1322: 

1323: Another way to describe this interface is that we color 

1324: the counterclockwise arc of $\p D$  from $p_0$ to

1325: $\zeta$ white and the clockwise arc from $p_0$ to

1326: $\zeta$ black, and $\beta$ then is the common boundary

1327: component between white and black in $\closure D$

1328: starting at $p_0$.

1329: We will call this type of interface a {\bf chordal interface},

1330: to differentiate it from the interface that we will later need

1331: when discussing the annulus crossing event.

1332: (The chordal interface was proved by Smirnov~\cite{Smirnov1} to converge to chordal

1333: SLE(6).)

1334: 

1335: 

1336: \proofof{Theorem~\ref{t.square}}

1337: We mostly concentrate on the case of site percolation on the

1338: triangular grid. The details in the case of bond percolation

1339: on $\Z^2$ are essentially the same.

1340: 

1341: The algorithm proceeds as follows. (See Figure~\ref{f.beta}.)

1342: There are four distinguished boundary arcs of $D$,

1343: which we call \lq\lq left\rq\rq, \lq\lq right\rq\rq, \lq\lq up\rq\rq\

1344: and \lq\lq down\rq\rq.

1345: Pick uniformly at random 

1346: an edge $e_0$ on the right hand boundary of 

1347: $D$, and let $p_0$ be its midpoint.

1348: Let $\zeta$ be the union of the top and left boundary

1349: segments of $D$.

1350: Explore the interface $\beta$ from $p_0$ to $\zeta$, examining the

1351: bits associated to sites in hexagons touching that interface,

1352: only as needed to continue with the determination of the interface.

1353: Note that the knowledge of $\beta$ suffices to determine if there

1354: is an open crossing from the right boundary of $D$ above $p_0$ to

1355: the left boundary of $D$: there is such a crossing if and only if

1356: $\beta$ terminates on the left boundary of $D$.

1357: 

1358: \begin{figure}

1359: \centerline{\epsfysize=2in%

1360: \SetLabels

1361: \L(1*.43)$\beta$\\

1362: \L(.95*.6)$p_0$\\

1363: \endSetLabels

1364: %\ShowGrid

1365: \AffixLabels{\epsfbox{beta.eps}}%

1366: \qquad

1367: \SetLabels

1368: \L(1.01*.25)$\beta'$\\

1369: \L(.95*.6)$p_0$\\

1370: \endSetLabels

1371: %\ShowGrid

1372: \epsfysize=2in%

1373: \AffixLabels{\epsfbox{betap.eps}}%

1374: }

1375: \begin{caption} {\label{f.beta}The interfaces $\beta$ and $\beta'$.}

1376: \end{caption}

1377: \end{figure}

1378: 

1379: Now let $\zeta'$ be the union of the bottom and left boundaries

1380: of $D$, and let $\beta'$ be the interface from $p_0$ to $\zeta'$

1381: that corresponds to the configuration $\omega'$ obtained by

1382: flipping all the colors of the hexagons in $\closure {D}$

1383: (alternatively, $\beta'$ is an interface that has black on the right

1384: and white on the left).

1385: Then $\beta'$ determines the existence of an open crossing from the

1386: right boundary below $p_0$ to the left boundary.

1387: Consequently, after the algorithm examines $\beta$ and $\beta'$,

1388: the correct value of $Q$ is determined.

1389: We now need to bound the revealment of this algorithm.

1390: 

1391: \begin{lemma}\label{l.Hdist}

1392: Let $\p_W$ denote the counterclockwise arc of $\p D\setminus\zeta$

1393: from $p_0$ to $\zeta$ (the white arc), and let

1394: $\p_B$ denote the clockwise arc of

1395: $\p D\setminus \zeta$ from $p_0$ to $\zeta$.

1396: Let $H$ be a grid hexagon in $\closure D$.

1397: Set $r_1:=\min\{\dist(H,\p_W),\dist(H,\p_B)\}$

1398: and $r_2:=\max\{\dist(H,\p_W),\dist(H,\p_B)\}$.

1399: Then

1400: $$

1401: \Pb{\p H\cap\beta\ne\emptyset \md p_0}

1402: \le \alpha_2(r_1)\,\alpha^+\bigl(2\,r_1+O(1),r_2-r_1-O(1)\bigr)\,.

1403: $$

1404: \end{lemma}

1405: \proof

1406: Suppose that $\p H\cap\beta\ne\emptyset$.

1407: Then there is a path in black hexagons from $\p H$

1408: to $\p_B$ and there is a path in white

1409: hexagons from $\p H$ to $\p_W$,

1410: because the chains of hexagons along the two

1411: sides of $\beta$ provide such paths.

1412: Suppose, for example, that

1413: $r_1=\dist(H,\p_W)$.

1414: Let $w$ be a closest point to $H$ in $\p_W$.

1415: Let $M$ be a half space that contains $D$,

1416: which satisfies $\dist(\p M,w)\le C$ where $C=O(1)$.

1417: (Here, we are using the fact that $D$ approximates a convex domain.)

1418: Let $w'$ be a point closest to $H$ on $\p M$.

1419: Then $\dist(w',H)\le r_1+C$.

1420: Set $R_1=2\,r_1+2\,C+2\diam(H)$ and $ R_2=r_2-r_1-C-\diam(H)$,

1421: and assume for now that $R_2>R_1$.

1422: Consider next the annulus centered at

1423: $w'$ with inner radius $R_1$ and outer radius $ R_2$.

1424: Now, $M$ intersected with this annulus

1425: contains a black crossing between the two boundary circles of this

1426: annulus (because there is a black crossing from

1427: $H$ to $\p_B$),

1428: and there are black and white crossings between

1429: $H$ and the circle of radius $r_1$ around the center

1430: of $H$.  These events are independent given $p_0$,

1431: which implies

1432: the lemma in the case $r_1=\dist(H,\p_W)$, $R_2>R_1$.

1433: If $R_2\le R_1$ and

1434: $r_1=\dist(H,\p_W)$, we only need to consider the

1435: crossings between $H$ and the circle of radius

1436: $r_1$ around its center.

1437: The case

1438: $r_1=\dist(H,\p_B)$ is treated similarly.

1439: \QED

1440: 

1441: \proofcont{Theorem~\ref{t.square}}

1442: Fix a hexagon $H\subset\closure D$.

1443: Note that the value of $r_1$ in the lemma does not depend on

1444: $p_0$, since $r_1=\dist(H,\p D\setminus\zeta)$.

1445: Let $z_0$ be the closest point to $H$ on the right boundary of

1446: the square which $D$ approximates.

1447: Observe that $r_2\ge |p_0-z_0|-O(1)$.

1448: This implies that for every $r\in[1,R]$, $\Pb{r/2\le r_2<r}\le O(r/R)$.

1449: Using the monotonicity of $\alpha^+$, Lemma~\ref{l.Hdist} therefore gives

1450: \begin{multline*}

1451: \Pb{\p H\cap\beta\ne\emptyset }

1452: \\

1453: \le O(1)\, \max_{0\le r_1\le R}\Bigl(

1454: \alpha_2(r_1)

1455: \sum_{j=0}^{\lceil \log_2 R\rceil}2^{-j}

1456: \alpha^+\bigl(2\,r_1+O(1),2^{-j}R-r_1-O(1)\bigr)

1457: \Bigr).

1458: \end{multline*}

1459: The same estimate also applies to $\beta'$, by symmetry.

1460: Consequently,~\eref{e.expostrong} gives

1461: \begin{multline*}

1462: \Pb{H\text{ examined by algorithm}}

1463: \\

1464: \le R^{o(1)}\max_{0\le r_1\le R}\Bigl(

1465: {r_1}^{-1/4}

1466: \sum_{j=0}^{\lceil \log_2 R\rceil}2^{-j}

1467: (2^{-j}R/r_1)^{-1/3}

1468: \Bigr)

1469: \\

1470: \le R^{o(1)}\max_{0\le r_1\le R}\bigl( {r_1}^{1/12} R^{-1/3} \bigr)

1471: = R^{-1/4+o(1)}\,,

1472: \end{multline*}

1473: as $R\to\infty$.

1474: This proves the theorem in the case of the triangular grid.

1475: 

1476: The proof for the square grid is essentially the same.

1477: Since in that case, we cannot use the values of the

1478: critical exponents, we just use the

1479: bounds $\alpha^+(r,r')\le C\,(r/r')^{\eps_0}$

1480: and $\alpha_2(r_1)\le C\, r_1^{-\eps_0}$, which are

1481: valid for some constants $C,\eps_0>0$.

1482: The theorem follows.

1483: \QED

1484: 

1485: 

1486: We now give the proof of Theorem~\ref{th:crossingquant}. We will not prove

1487: Theorem~\ref{th:crossingquantsquare}, but rather simply say that it is 

1488: proved in a similar way.

1489: 

1490: \proofof{Theorem~\ref{th:crossingquant}}

1491: Fix $\gamma < 1/8$.

1492: Let $f$ be the indicator function $f(\omega)=1_{A_m'}(\omega)$.

1493: By (\ref{e.page14}), we have

1494: $$

1495: N(A_m',m^{-\gamma})= \sum_{\emptyset\neq S \subseteq [n_m]} \hat{f}(S)^2 (1-2\,m^{-\gamma})^{2|S|}

1496: $$

1497: where $n_m$ is the number of sites in $D$.  By Corollary~\ref{c.square},

1498:  with $\epsilon >0$ chosen so that $2\,\gamma+\eps < 1/4$,

1499: this is at most

1500: \begin{multline*}

1501: C

1502: \sum_{k=1}^{n_m} (1-2m^{-\gamma})^{2k}\, k\,m^{-1/4+\eps}\\

1503: \le C\,{m^{-1/4+\epsilon}}

1504: \sum_{k=1}^{\infty}k\, (1-2\,m^{-\gamma})^{\frac{m^\gamma}{2}\frac{4k}{m^\gamma}} 

1505: \le C\,{m^{-1/4+\epsilon}}

1506: \sum_{k=1}^{\infty} k\, e^{-4k{m^{-\gamma}}} 

1507: \end{multline*}

1508: It is easy to check that

1509: $$

1510: \sum_{k=1}^{\infty} k\, e^{-4k{m^{-\gamma}}} \le O(m^{2\gamma})\,,

1511: $$

1512: and so the result follows immediately.

1513: \QED

1514: 

1515: \subsection{Annulus case}\label{ss.ann}

1516: 

1517: For the proof of Theorem~\ref{th:except}, we will need 

1518: the following variant of Theorem~\ref{t.square}

1519: regarding the percolation crossings of an annulus.

1520: 

1521: 

1522: \begin{theorem}\label{t.annulus}

1523: Let $2\le r< R$. 

1524: Let $\croRr$ be the indicator function for the 

1525: event that there is a crossing of the annulus

1526: $\{z\in\R^2: r\le z\le R\}$

1527: from the inner circle to the outer circle

1528: by a cluster of white hexagons.

1529: Then there is a randomized algorithm determining $\croRr$ with

1530: \begin{equation}\label{e.td}

1531: \delta\le r^{o(1)}\,\alpha(r,R)\,\alpha_2(r)\,.

1532: \end{equation}

1533: \end{theorem}

1534: 

1535: We stress that the $r^{o(1)}$ factor depends on $r$ only and not on $R$.

1536: It is possible to replace the factor $r^{o(1)}$ by $O(1)$,

1537: but in order to do this it seems that one would first need to

1538: appeal to the analogue of~\eref{e.multip} for $\alpha_2$, which

1539: is proved in the appendix.

1540: In order to have a more direct proof of our main theorem,

1541: we prefer, at this point, not to rely on the appendix.

1542: By~\eref{e.onearmstrong} and~\eref{e.expo},

1543: the right hand side in~\eref{e.td} is equal to

1544: $R^{-5/48+o(1)}\,r^{-7/48}$, but its writing in~\eref{e.td} is more

1545: suggestive and more useful.

1546: 

1547: 

1548: 

1549: \medskip

1550: 

1551: Since $\|\croRr\|^2=\alpha(r,R)$, 

1552: Theorems~\ref{t.noise} and~\ref{t.annulus} give 

1553: 

1554: \begin{corollary}\label{c.croRr}

1555: $$

1556: \sum_{|S|=k}\hat\croRr(S)^2

1557: \le 

1558: k\,r^{o(1)}\,\alpha(r,R)^2\,\alpha_2(r)

1559: $$

1560: holds when $1\le r<R<\infty$ and $k>0$.

1561: \QED

1562: \end{corollary}

1563: 

1564: Before we prove Theorem~\ref{t.annulus},

1565: we have to discuss the kind of interface that is used by the algorithm,

1566: as it is slightly different from the interface used to determine

1567: a possible crossing of a square.

1568: 

1569: Fix $R>0$ large.

1570: Let $\closure D=\closure D_R$

1571: be the union of the hexagons of the hexagonal grid 

1572: that intersect the disk $|z|\le R$.

1573: Let $V^*=V^*_R$ denote the set of vertices of the hexagonal grid

1574: that are in $\closure D$.

1575: Let $p_0$ be some point in $\p \closure D\setminus V^*$.

1576: Let $H_0$ denote the hexagon containing the origin, and

1577: let $q_0$ be some point in $\p H_0\setminus V^*$.

1578: We define the {\bf radial interface} $\beta=\beta(R,p_0,q_0,\omega)$,

1579: inductively as a simple path from $p_0$ to $q_0$.

1580: (See Figure~\ref{f.annulusinterface}.)

1581: The construction is segment by segment, and the concatenation

1582: of the first $m$ segments will be denoted $\beta_m$.

1583: If the (unique) hexagon in $\closure D$

1584: containing $p_0$ is white [respectively, black], then 

1585: the first segment $\beta_1$ of $\beta$ traverses the boundary

1586: of that hexagon [counter-] clockwise 

1587: until the first encounter with a point in $V^*$.

1588: Suppose inductively $\beta_m$ has been constructed,

1589: that it is a simple path, that

1590: $p_m\in V^*$ and $p_0$ are the two

1591: endpoints of $\beta_m$,

1592: and that there is a path $\alpha$ in

1593: the hexagonal grid in $\closure D$ from

1594: $p_m$ to $q_0$ whose only intersection with $\beta_m$ is $p_m$.

1595: The first step of such a path $\alpha$ is along an edge $\hat e$

1596: starting at $p_m$. If there is just one possible $\hat e$ among all

1597: such $\alpha$, then $\beta_{m+1}$ also uses that

1598: edge $\hat e$. Clearly, there are at most two possible $\hat e$,

1599: since the edge terminating at $p_m$ and used by $\beta_m$

1600: cannot be used. If there are two possible $\hat e$, then

1601: $\beta_{m+1}$ chooses between them according to the color of

1602: the hexagon containing them both; i.e., the hexagon just encountered

1603: by $\beta_m$. If that hexagon is white [respectively, black],

1604: then the edge chosen is the one that traverses $H$

1605: [counter-] clockwise. If the edge chosen contains $q_0$,

1606: then the path stops at $q_0$ and the construction terminates.

1607: Otherwise, $\beta_{m+1}$ is defined as the union of $\beta_m$ and

1608: the chosen continuation edge. This completes the definition of $\beta$.

1609: 

1610: \begin{figure}

1611: \SetLabels

1612: (.503*.50)$q_0$\\

1613: \T(.44*.01)$p_0$\\

1614: \B(.95*.61)$\beta^T$\\

1615: \B(.95*.43)$\beta\setminus\beta^T$\\

1616: \endSetLabels

1617: %\ShowGrid

1618: \centerline{\epsfysize=2.5in%

1619: \AffixLabels{%

1620: \epsfbox{annular.eps}}%

1621: }

1622: \begin{caption} {\label{f.annulusinterface}The radial interface

1623: $\beta$.}

1624: \end{caption}

1625: \end{figure}

1626: 

1627: 

1628: It is not hard to verify that for every simple path $\hat\beta$

1629: in the hexagonal grid from $p_0$ to $q_0$ that stays in

1630: $\closure D$, the probability that $\beta=\hat\beta$

1631: is precisely $2^{-n}$ if $n$ is the number of

1632: hexagons in $\closure D$ that intersect $\hat\beta$.

1633: However, we will not use this fact.

1634: 

1635: Let $r\in[0,R]$.

1636: We now define a truncated version of $\beta$, which will

1637: suffice, as we will see, to determine $\croRr$.

1638: We say that $\beta$ completed a [counter-] clockwise loop

1639: at some dual vertex $v\in V^*$ if $v$ is visited by $\beta$

1640: and there is a hexagon $H$ containing $v$ and  another  point

1641: $u\in\p H$, which was visited by $\beta$ prior to

1642: $v$, 

1643: and the oriented arc of $\beta$ from $u$ to $v$ together with the

1644: line segment $[v,u]\subset H$ form a [counter-] clockwise

1645: loop surrounding $0$.

1646: Let $\beta^T$ denote the initial segment of 

1647: $\beta$ up to the first time in which $\beta$ completed

1648: a counterclockwise loop around $0$ or until it hits

1649: $q_0$, if there is no such loop.

1650: 

1651: 

1652: \begin{lemma}\label{l.betadet}

1653: The truncated radial interface $\beta^T$ meets 

1654: the disk $|z|\le r$ if and only if $\croRr=1$.

1655: \end{lemma}

1656: \proof

1657: Let $[u,v]$ be an edge in $\beta^T$, with $u$ occuring

1658: before $v$ along $\beta$.

1659: We claim that if the hexagon $H$ immediately to the right

1660: of $[u,v]$ is contained in $\closure D$, then it is white.

1661: Indeed, suppose the contrary. Let $w$ be the first vertex along 

1662: $\beta$ in which $\p H$ is visited, and let

1663: $\beta^w$ be the initial segment of $\beta$ from $p_0$

1664: to $w$.  Observe that the counterclockwise arc from 

1665: $w$ to $v$ is a feasible continuation of $\beta^w$,

1666: since $\beta$ contains a path from $v$ to $q_0$ and

1667: there is no other point but $w$ in

1668: $\p H\cap\beta^w$.

1669: Since we are assuming that $H\subset\closure D$ is black,

1670: it follows that the immediate continuation of $\beta^w$

1671: was along $\p H$ in the counterclockwise direction until

1672: some $w'\in V^*$ is hit.

1673: 

1674: Consider the directed cycle obtained by joining the line segment

1675: $[u,w']$ to the arc of $\beta$ from $w'$ to $u$.

1676: This directed cycle surrounds $v$, because $w$ is connected

1677: in $\beta^w$ to $p_0$, which is certainly in the unbounded

1678: component of this cycle, and the line segment

1679: $[v,w]$ intersects the cycle precisely once,

1680: crossing the line segment $[u,w']$ inside $H$.

1681: Moreover, if we consider the orientation in which

1682: these two line segments cross, we conclude that

1683: the cycle surrounds $v$ counterclockwise.

1684: Because the arc of $\beta$ from $v$ to $q_0$

1685: does not intersect the cycle, we conclude that the cycle also

1686: surrounds $0$ counterclockwise.

1687: This contradicts the definition of the truncated path $\beta^T$,

1688: since we are assuming $v\in\beta^T$.

1689: This verifies our claim, that to the right of edges in

1690: $\beta^T$ are only white hexagons and hexagons that are not

1691: contained in $\closure D$.

1692: 

1693: Note that if $e$ and $e'$ are two consecutive segments

1694: along $\beta$, then the hexagon to the right of $e$ is

1695: either the same as the one to the right of $e'$, or these

1696: hexagons are adjacent. We therefore conclude that every

1697: white hexagon visited by $\beta^T$ is connected by

1698: a chain of white hexagons to $\p D$.

1699: Therefore, if $\beta^T$ hits the set $|z|\le r$,

1700: then clearly $\croRr=1$.

1701: 

1702: Now suppose that $\beta^T$ does not hit $|z|\le r$.

1703: This implies that $\beta^T$ has terminated by completing a

1704: counterclockwise loop around the set $|z|\le r$.

1705: Consider a hexagon $H$ on the inner boundary of this loop.

1706: Because the orientation of the loop is

1707: counterclockwise, the first time in which $H$ is

1708: visited, the path chose to traverse $\p H$ counterclockwise.

1709: This implies that the hexagon is black.

1710: Thus, there is a loop of black hexagons in $\closure D$

1711: that surrounds the set $|z|\le r$. This implies that $\croRr=0$.

1712: \QED

1713: 

1714: 

1715: We can now specify the algorithm promised by Theorem~\ref{t.annulus}.

1716: The algorithm starts by selecting the point $p_0$ uniformly along $\p\closure D$,

1717: and selecting $q_0$ arbitrarily in $\p H_0\setminus V^*$.

1718: It then proceeds to inspect the colors of the hexagons necessary to develop

1719: the truncated interface $\beta^T$, until it terminates or

1720: hits the set $|z|\le r$. At that point, the correct value

1721: of $\croRr$ is determined, by Lemma~\ref{l.betadet}.

1722: 

1723: In order to bound the revealment of this algorithm,

1724: it will be convenient to introduce a different interface, which in the end will

1725: turn out to be equivalent to $\beta$.

1726: 

1727: Let $\hat D$ denote the branched double cover of $\closure D$ about $0$,

1728: and let $\phi:\hat D\to \closure D$ denote the projection map.

1729: Concretely, define $\hat D$ as the preimage of

1730: $\closure D$ under the map $\phi(z)=z^2$.

1731: Let $\hat p_0$ be one of the preimages of $p_0$ under $\phi$,

1732: and let $\hat q_0$ be one of the preimages of $q_0$.

1733: Let $\hat H_0$ be the closure of one of the connected components

1734: of $\phi^{-1}(H_0)\setminus [\hat q_0,-\hat q_0]$.

1735: Let $\mfh$ denote the set of hexagons $H$ that are contained

1736: in $\closure D$.

1737: Let $\hmfh$ denote the set of connected components

1738: of preimages $\phi^{-1}(H)$, $H\in\mfh$, except that

1739: the single preimage of $H_0$ is replaced by

1740: the two sets $\hat H_0$ and $-\hat H_0$.

1741: Note that if $\hat H\in\hmfh$,

1742: then $-\hat H\in\hmfh$ and $\phi(\hat H)=\phi(-\hat H)\in\mfh$.

1743: Let $\hmfh' \subset \hmfh $ be a maximal collection

1744: of elements of $\hmfh$ with the property that

1745: $\hmfh'\cap \{-\hat H:\hat H\in\hmfh'\}=\emptyset$.

1746: Now color at random each of the elements of $\hmfh'$ white or

1747: black independently, with probability $1/2$.

1748: For every $\hat H\in\hmfh'$, let $-\hat H$ have the opposite

1749: color to the color of $\hat H$.

1750: 

1751: Now let $\hat \beta$ denote the chordal interface in $\hat D$

1752: from $\hat p_0$ to $-\hat p_0$, with white cells on the right

1753: and black cells on the left, as defined in the simply connected setting

1754: in Subsection~\ref{ss.simply}.

1755: That is, we consider the exterior of the counterclockwise arc from $\hat p_0$

1756: to $-\hat p_0$ along $\p \hat D$ as white, the exterior of the complementary

1757: arc as black, and take $\hat \beta$ as the interface between white and

1758: black passing through $\hat p_0$ and through $-\hat p_0$.

1759: Finally, let $\beta^\dagger:= \phi(\hat\beta)\setminus \interior(H_0)$.

1760: 

1761: \begin{lemma}

1762: Given $p_0$ and $q_0$,

1763: the laws of $\beta^\dagger$ and of $\beta$ are the same.

1764: \end{lemma}

1765: (In this statement, we consider $\beta$ as a set, and forget

1766: about the fact that it has the structure of an oriented path.)

1767: 

1768: \proof

1769: The map $z\mapsto -z$ preserves $\hat\beta$, by the symmetry

1770: of the interface. Consequently, near every point $p\in \beta^\dagger\setminus \{p_0,q_0\}$,

1771: $\beta^\dagger$ looks like a piecewise linear path. Moreover,

1772: $\beta^\dagger$ is connected and contains $p_0$. Since a compact simple path has two endpoints,

1773: we conclude that $q_0\in\beta^\dagger$ as well.

1774: We now consider building $\beta^\dagger$ by adding one segment at a time.

1775: When it hits a previously unvisited hexagon $H$ which is contained in $\closure D$,

1776: it is equally likely (given its past) to turn right or left.

1777: (This is because both preimages of $H$ are unvisited by both preimages of the

1778: past of $\beta^\dagger$.) When it hits a previously visited hexagon (or

1779: a hexagon that is not contained in $\closure D$), it turns in such a way

1780: that it will eventually be able to reach $q_0$ without crossing itself,

1781: and this uniquely specifies this turn. Consequently, the lemma follows.

1782: \QED

1783: 

1784: \begin{remark}

1785: The radial interface converges to radial SLE(6) as the mesh tends to zero.

1786: \end{remark}

1787: 

1788: 

1789: 

1790: 

1791: 

1792: \proofof{Theorem~\ref{t.annulus}}

1793: Given all of our preparations, the proof is rather easy.

1794: We have shown that the above algorithm provides the correct answer.

1795: It therefore remains to estimate its revealment.

1796: Consider some hexagon $H\subset\closure D$.

1797: We want to prove that the right hand side of~\eref{e.td}

1798: is an upper bound for the probability that $H$ is examined.

1799: Let $a:=\dist(0,H)$, $b:=\dist(H,\p\closure D)$ and $c:=\dist(p_0,H)$.

1800: Let $S_1$ be the disk of radius $(a\wedge b)/2$ concentric with $H$,

1801: and let $\hat S_1$ be one of the two connected components

1802: of $\phi^{-1}(S_1)$.

1803: We have to bound the probability that the algorithm inspects $H$.

1804: Clearly, we may assume $a\ge r-O(1)$.

1805: For $H$ to be inspected,

1806:  $\beta^T$ has to get to the circle

1807: $|z|=R\wedge (2\,a)$.  This probability

1808: is $\alpha(2\,a,R)$.

1809: Given that this has happened, how can we estimate the

1810: probability that $\beta$ is adjacent to $H$?

1811: At this point, we use the equivalence of

1812: $\beta$ and $\beta^\dagger$.

1813: The information that $\beta^\dagger$ reached the circle

1814: $|z|=R\wedge (2\,a)$ bears no impact on the distribution

1815: of the colors of the cells in $\hmfh$ whose images under

1816: $\phi$ intersect $S_1$. (Here, we assume that $a$ is not too small,

1817: so that the corresponding sets of cells are disjoint.

1818: Certainly $a>10$ suffices. If $a$ is smaller,

1819: then the estimate we are now striving for is trivial.)

1820: Since there is no hexagon intersecting both $\hat S_1$ and $-\hat S_1$,

1821: it follows that the conditional distribution of the

1822: colors of the cells meeting $\hat S_1$ is uniform

1823: i.i.d. Consequently, the conditional probability that

1824: $\beta^\dagger$ hits $H$ is bounded by $\alpha_2((a\wedge b)/2)$.

1825: Thus,

1826: $$

1827: \Pb{H\text{ visited}}\le

1828: O(1)\,\alpha(2\,a,R)\,

1829: \alpha_2((a\wedge b)/2)\,.

1830: $$

1831: In the case $b\ge a\ge 2\,r$, we may use independence on disjoint

1832: sets to conclude that 

1833: \begin{equation*}

1834: \begin{aligned}

1835: &\Pb{H\text{ visited}}

1836:  \\

1837: &\qquad

1838: \le

1839: O(1)\, \alpha_2(r)\,\alpha_2(2\,r,a/2)\,\alpha(2\,a,R)

1840: \le

1841: O(1)\,

1842: \alpha_2(r)\,\alpha(2\,r,a/2)\,\alpha(2\,a,R)

1843: \\

1844: &\qquad

1845: \overset{\eref{e.rsw}}{\le}

1846: O(1)\,

1847: \alpha_2(r)\,\alpha(r,2\,r)\,\alpha(2\,r,a/2)\,

1848: \alpha(a/2,2\,a)\,\alpha(2\,a,R)

1849: \\

1850: &\qquad

1851: \overset{\eref{e.multip}}{\le}

1852: O(1)\,

1853: \alpha_2(r)\,\alpha(r,R)\,.

1854: \end{aligned}

1855: \end{equation*}

1856: On the other hand, if $b\ge a$ and $a\le 2\,r$,

1857: then we use our assumption $a\ge r-O(1)$ and~\eref{e.expo}

1858: to get

1859: $\alpha_2(a/2)\le r^{-1/4+o(1)}\le\alpha_2(r)\,r^{o(1)}$,

1860: which is also sufficient.

1861: 

1862: In the case $b< a$, a similar argument 

1863: (and similar to the proof of Lemma~\ref{l.Hdist})

1864: shows that

1865: $$%\begin{multline*}

1866: \Pb{H\text{ visited }\md c}\le O(1)\,\alpha_2(b/2)\,\alpha^+\bl(2\,b+O(1),c-b-O(1)\br).

1867: $$%\end{multline*}

1868: Next, picking a constant $q\in(1/4,1/3)$, we then have by the above and~\eref{e.expostrong}

1869: $$%\begin{multline*}

1870: \Pb{H\text{ visited }\md c}

1871: \le O(1)\, \alpha_2(b/2)\, (c/b)^{-q}.

1872: $$%\end{multline*}

1873: As in the proof of Theorem~\ref{t.square}, we have $\Pb{2^j\le c<2^{j+1}} \le O(1)\,2^j/R$.

1874: It easily follows that

1875: $$

1876: \Pb{H\text{ visited}}\le

1877: O(1) \,\alpha_2(b/2)\, (R/b)^{-q}.

1878: $$

1879: If $b/2>r$, then we may estimate

1880: $$

1881: \alpha_2(b/2) \le \alpha_2(r)\, \alpha_2(2\,r,b/2) \le \alpha_2(r)\, \alpha(2\,r,b/2)

1882: $$

1883: and 

1884: $$

1885: (R/b)^{-q}

1886: \overset{\eref{e.expostrong}}{\le}

1887: O(1)\, \alpha(b/2,R)

1888: $$

1889: and we get from~\eref{e.multip} and the above

1890: $$

1891: \Pb{H\text{ visited}}\le O(1)\, \alpha_2(r)\, \alpha(r,R)\,.

1892: $$

1893: If $b/2\le r$, we use instead

1894: $$ 

1895: \alpha_2(b/2)\, (R/b)^{-q}\le O(1)\, \alpha_2(b/2)\, \alpha_2(b/2,r)\, \alpha(r,R)

1896: \overset{\eref{e.expostrong}}{\le}

1897: r^{o(1)}\, \alpha_2(r)\, \alpha(r,R)\,.

1898: $$

1899: This completes the proof.

1900: \QED

1901: 

1902: 

1903: \begin{remark}

1904: It is easy to see that if we assume the analogue of~\eref{e.multip} for 

1905: $\alpha_2$ proved in the appendix,

1906: then the $r^{o(1)}$ term in~\eref{e.td} can be replaced by $O(1)$.

1907: \end{remark}

1908: 

1909: \section{Exceptional times}\label{sec:except}

1910: 

1911: In this section we prove Theorem \ref{th:except}. 

1912: We point out that the absolute key necessary step is to get a good

1913: bound on the correlation for an event occurring at two different but

1914: close by times. Once this is done, the rest is fairly standard.

1915: Proposition A16 in Lawler~\cite{Lawler}

1916: indicates this general type of argument.

1917: 

1918: \proofof{Theorem~\ref{th:except}}

1919: By Kolmogorov's 0-1 law, it suffices to prove that with positive probability

1920: there are times in $[0,1]$ when the origin is in an infinite cluster.

1921: Fix $R>2$ large and let $V_{t,R}$ be the event

1922: that at time $t$ there is an open path from the origin to distance $R$ away.

1923: We then let

1924: $$

1925: X=X_R:= \int_0^1 1_{V_{t,R}} \, dt 

1926: $$

1927: be the Lebesgue measure of the set of times in $[0,1]$ at which $V_{t,R}$ occurs.

1928: The first moment of $X$ is given by

1929: $$

1930: \Eb{X}=\int_0^1 \Pb{V_{t,R}}\,dt=\Pb{V_{0,R}} =\alpha(R)\,.

1931: $$

1932: The second moment is

1933: \begin{equation}\label{e.2nd}%\begin{multline*}

1934: \Eb{X^2}=

1935: \EB{\int_0^1\int_0^1 1_{V_{s,R}}\,1_{V_{s',R}}\,ds\,ds'}

1936: =

1937: \int_0^1\int_0^1 \Pb{V_{s,R}\cap V_{s',R}}\,ds\,ds'.

1938: \end{equation}%\end{multline*}

1939: 

1940: 

1941: For each site $v$ we let 

1942: $$

1943: \chi_v^s:=\begin{cases} -1&v \text{ is open at time } s\\

1944: 1&\text{otherwise},

1945: \end{cases}

1946: $$

1947: and for a finite set of sites $S$ set

1948: $$

1949: \chi_S^s:=\prod_{v\in S}\chi_v^s\,.

1950: $$

1951: Fix $s,s'\in[0,1]$, and set $t:=|s-s'|$.

1952: Recall that the state of a site $v$ flips between closed and open

1953: with rate $1/2$. Equivalently, we may think of the state as being

1954: re-randomized with rate $1$.

1955: Consequently, $\Pb{\chi_v^{s'}=\chi_v^s \md \omega_s} = e^{-t}+(1-e^{-t})/2

1956: =(1+e^{-t})/2$,

1957: and hence,

1958: $$

1959: \Eb{\chi_v^s\,\chi_v^{s'}}=\exp(-t)

1960: \,,\qquad

1961: \Eb{\chi_S^{s}\,\chi_S^{s'}}=\prod_{v\in S}\Eb{\chi_v^s\,\chi_v^{s'}}=\exp(-t\,|S|)\,.

1962: $$

1963: Moreover, if $S\ne S'$, then $\Eb{\chi_S^s\,\chi_{S'}^{s'}}=0$.

1964: Consequently, if $f$ is a function depending on the states of finitely many lattice

1965: points and has the expansion

1966: $f(\omega)=\sum_S\hat f(S)\,\chi_S(\omega)$, then

1967: \begin{equation}\label{e.timeexpansion}

1968: \Eb{f(\omega_s)\,f(\omega_{s'})}=

1969: \sum_S\hat f(S)^2 \exp(-t\,|S|)\,.

1970: \end{equation}

1971: 

1972: Let $f_r^R(\omega)$ be as in Theorem~\ref{t.annulus}.

1973: Fix some $t\in (0,1]$ and let $r\in [2,R)$.

1974: Clearly, $0\le f_0^R(\omega)\le f_0^r(\omega)\,f_{2r}^R(\omega)$ for every $\omega$.

1975: Consequently,

1976: \begin{multline*}

1977: \Pb{\Vss}=

1978: \Eb{f_0^R(\omega_s)\,f_0^R(\omega_{s'})}

1979: \le

1980: \Eb{f_0^r(\omega_s)\,f_{2r}^R(\omega_s)\,

1981: f_0^r(\omega_{s'})\,f_{2r}^R(\omega_{s'})}

1982: \\

1983: =

1984: \Eb{f_0^r(\omega_s) \, f_0^r(\omega_{s'})}

1985: \Eb{ f_{2r}^R(\omega_s)\,f_{2r}^R(\omega_{s'})}

1986: \le

1987: \Eb{f_0^r(\omega_s) } \Eb{ f_{2r}^R(\omega_s)\,f_{2r}^R(\omega_{s'})} .

1988: \end{multline*}

1989: (To obtain the second equality, we have used the independence on 

1990: disjoint sets of sites.) Thus,

1991: $$

1992: \Pb{\Vss}\le

1993: \alpha(r)\,

1994: \Eb{ f_{2r}^R(\omega_s)\,f_{2r}^R(\omega_{s'})}

1995: =\alpha(r)\, \sum_S e^{-t|S|}\hat f_{2r}^R(S)^2.

1996: $$

1997: The latter sum restricted to $S$ with $|S|=k$ for fixed $k\ne 0$

1998: is estimated using Corollary~\ref{c.croRr},

1999: while for $k=0$, we use $\hat f_{2r}^R(\emptyset)=\alpha(2r,R)$. This yields

2000: $$

2001: \Pb{\Vss}\le

2002: \alpha(r)\,\Bigl(\alpha(2\,r,R)^2+

2003: r^{o(1)}\,\sum_{k=1}^\infty 

2004: e^{-kt}

2005: k\,\alpha(2\,r,R)^2\, \alpha_2(r)\Bigr).

2006: $$

2007: It is easy to check that $\sum_{k=1}^\infty k\,e^{-kt}\le O(t^{-2})$.

2008: This and the inequalities~\eref{e.multip} and~\eref{e.rsw} allow

2009: us to write this estimate as

2010: \begin{equation}\label{e.zz}

2011: \Pb{\Vss}\le

2012: O(1)\,

2013: \alpha(R)^2\,\alpha(r)^{-1}\bl(1+r^{o(1)}\,t^{-2}\,\alpha_2(r)\br)\,.

2014: \end{equation}

2015: We proved the above claim for all 

2016: $t\in (0,1]$ and $r\in [0,R)$ but

2017: now we observe that~\eref{e.zz} is also trivially true

2018: when $r \ge R$ as well. 

2019: We now choose $r=2\,t^{-8}=2\,|s-s'|^{-8}$.

2020: Applying this in~\eref{e.zz} with~\eref{e.onearm} and~\eref{e.expo} gives

2021: \begin{equation}\label{e.uselater}

2022: \Pb{V_{s,R}\cap V_{s',R}}

2023: \le O(1)\,\alpha(R)^2\, |s-s'|^{-5/6+o(1)}.

2024: \end{equation}

2025: Hence,

2026: \begin{equation}\label{e.fin}

2027: \int_0^1\int_0^1\Pb{V_{s,R}\cap V_{s',R}}\,ds\,ds'

2028: \le O(1)\,\alpha(R)^2.

2029: \end{equation}

2030: 

2031: The Cauchy-Schwarz inequality tells us that

2032: $$

2033: \Pb{X>0} \ge 

2034: \frac{

2035: \Eb{X}^2

2036: }{

2037: \Eb{X^2}

2038: }\,.

2039: $$

2040: Consequently, the above inequality, the fact that $\Eb{X}=\alpha(R)$,

2041: the expression~\eref{e.2nd} for $\Eb{X^2}$ and~\eref{e.fin}

2042: show that $\inf_{R>0}\Pb{X_R>0}>0$.

2043: Let $T_R:=\{t\in[0,1]:V_{t,R}\text{ holds}\}$.

2044:  Fatou's lemma tells us that with positive probability

2045: $T_R\ne \emptyset$ for infinitely many $R\in\N$.

2046: Since $T_R\supset T_{R'}$ when $R'>R$,

2047: this implies that 

2048: $$

2049: \Pb{\cap_{R>0}\,\, \{T_R\ne\emptyset\}}>0.

2050: $$

2051: Our goal is to show that $\Pb{\bigcap_R T_R \neq \emptyset}>0$.

2052: Since the $T_R$'s are not closed sets,

2053: $\cap_{R>0}\,\, \{T_R\ne\emptyset\}$

2054: does not immediately imply $\bigcap_R T_R\ne\emptyset$.

2055: (The reason that $T_R$ is not necessarily closed is that the set of times at

2056: which an edge is open is not a closed set since we have a right continuous

2057: process.) This technicality is taken care of by the following lemma from \cite{HPS}. 

2058: 

2059: \begin{lemma}\label{l.HPS} (\cite{HPS})

2060: Let $0<p<1$ and let $G$ be any graph where $\pi_p(\calC)=0$.

2061: Let $\{\omega_t\}$ represent our dynamical percolation process in

2062: that $\omega_t(v)$ is the state of vertex $v$ at time $t$.

2063: Consider the process $\{\bar\omega_t\}$ obtained from $\{\omega_t\}$

2064: by setting, for every vertex $v$,

2065: the set  $\{t\, : \, \bar\omega_t(v) =1\}$ to be 

2066: the closure of the set $\{t\, : \, \omega_t(v) =1\}$.

2067: Then $\bPsi_p$-a.s., for every vertex $v$ we have

2068: $$

2069:   \{t \in [0, \infty) \: :\:  v \, 

2070: \mbox{ \rm percolates in } \bar\omega_t \} \, = \,

2071:  \{t \in [0, \infty) \: :\:  v \, \mbox{ \rm percolates in } \omega_t \} \,.

2072: $$

2073: In particular, a.s.\ this set of times is closed.

2074: \end{lemma}

2075: 

2076: Returning to our proof,

2077: let $\overline{T_R}$ be the closure of $T_R$. It is easily checked that

2078: $$

2079: \bigcap_{R>0}\overline{T_R}= 

2080:   \{t \in [0,1] \: :\:  0 \, 

2081: \mbox{ \rm percolates in } \bar\omega_t \},

2082: $$

2083: where $\{\bar\omega_t \}$ is defined in Lemma \ref{l.HPS}.

2084: By compactness, if the $T_R$'s are all nonempty, it follows that 

2085: $\bigcap_R\overline{T_R}$ is nonempty. This implies that there 

2086: is some time at which $\bar\omega_t$ percolates and hence by

2087: Lemma \ref{l.HPS}, some time at which the original process $\omega_t$ percolates.

2088: \QED

2089: 

2090: For future reference, we note that Lemma~\ref{l.HPS} implies that a.s.

2091: \begin{equation}\label{e.ctr}

2092: {\bigcap_{R>0}\overline{T_R}= 

2093: \bigcap_{R>0}{T_R} }.

2094: \end{equation}

2095: 

2096: 

2097: \section{Hausdorff dimension of exceptional times} \label{sec:hd}

2098: 

2099: In this section, we prove Theorem \ref{th.hd}. This is separated into two theorems,

2100: Theorem \ref{t.lowerbound} and Theorem \ref{t.upperbound}, where lower and upper bounds

2101: are given. We point out however that the lower bound is simply a refinement of

2102: the argument for proving that there exist exceptional times.

2103: First note that the fact that the  Hausdorff dimension is an almost sure

2104: constant follows immediately from ergodicity.

2105: 

2106: \begin{theorem}\label{t.lowerbound}

2107: A.s.,  the Hausdorff dimension of the set of exceptional times is at least $1/6$.

2108: \end{theorem}

2109: 

2110: \proof

2111: Fix $\gamma < 1/6$. It suffices by ergodicity and countable additivity

2112: to show that with positive probability,

2113: the set of exceptional times in $[0,1]$ at which the origin percolates

2114: has Hausdorff dimension at least $\gamma$.

2115: For each integer $R$, let as before $V_{t,R}$ be the event

2116: that at time $t$ there is a path from the origin to distance $R$ away and

2117: define a random measure $\sigma_R$ on $[0,1]$ by

2118: $$

2119: \sigma_R(S)=\frac{1}{\alpha(R)}\int_{S}1_{V_{t,R}}dt

2120: $$

2121: for each Borel set $S\subset [0,1]$.

2122: 

2123: The results in the previous section immediately give that

2124: $\E[\| \sigma_R \|] =1$ and $\E[\| \sigma_R \|^2] \le O(1)$

2125: where $\| \sigma_R \|$ denotes the total variation of the measure $\sigma_R$.

2126: 

2127: Cauchy-Schwarz gives

2128: $$

2129: \Eb{\|\sigma_R\|^2}^{1/2}\,\Pb{\|\sigma_R\|>1/2}^{1/2}

2130: \ge

2131: \Eb{\|\sigma_R\| 1_{\|\sigma_R\|>1/2}}

2132: \ge

2133: \Es{\|\sigma_R\| }-1/2=1/2.

2134: $$

2135: Consequently, 

2136: $\Pb{\| \sigma_R \| > 1/2} \ge  C_1 $

2137: for some constant $C_1 >0$.

2138: Given a measure $m$ on $[0,1]$ and $\gamma >0$, let

2139: $$

2140: {\cal E}_{\gamma} (m) = \int \!\!\int {|t-s|}^{- \gamma} \, dm (t) \,   dm(s) .

2141: $$

2142: Note that

2143: $$

2144: \Eb{{\cal E}_{\gamma} (\sigma_R)}

2145: =

2146: \EB

2147: {\int_0^1\!\int_0^1\frac{d\sigma_R (t) \,   d\sigma_R(s)}{ {|t-s|}^ \gamma}  }=

2148: \int_0^1\!\int_0^1

2149: \frac{\Pb{{V_{t,R}}\cap{V_{s,R}} }}{{\alpha(R)^2}\, |t-s|^\gamma}\, dt\,ds \,.

2150: $$

2151: Therefore, 

2152: by~\eref{e.uselater} and $\gamma<1/6$,

2153: $$

2154: C_2:=\sup_R

2155: \Eb{{\cal E}_{\gamma} (\sigma_R)}<\infty\,.

2156: $$

2157: 

2158: By Markov's inequality, for all $R$ and for all $T$,

2159: $$

2160: \Pb{{\cal E}_{\gamma} (\sigma_R)\ge  C_2T} \le 1/T.

2161: $$

2162: Choose $T$ so that $1/T < C_1/2$.

2163: Letting 

2164: $$

2165: U_R=\{\| \sigma_R \| > 1/2\}\cap \{{\cal E}_{\gamma} (\sigma_R)\le  C_2T\},

2166: $$

2167: by the choice of $T$, we have that 

2168: $$

2169: \Pb{U_R} \ge C_1/2.

2170: $$

2171: By Fatou's lemma,

2172: $$

2173: \Pb{\limsup_{R\to\infty} U_R} \ge C_1/2.

2174: $$

2175: We now show that on the event $\limsup_R U_R$, the Hausdorff dimension of the 

2176: set of percolating times in $[0,1]$ is at least $\gamma$. 

2177: Let $\overline{T_R}$ again be the closure of the set of times in $[0,1]$ at which 

2178: there is a path from the origin to distance $R$ away. Clearly $\sigma_R$ is supported on

2179: $\overline{T_R}$.

2180: By~\eref{e.ctr}, it suffices to prove that

2181: $\bigcap_{R>0}\overline{T_R}$ has Hausdorff dimension at least $\gamma$

2182: on the event $\limsup_R U_R$.

2183: This is achieved in the following

2184: (deterministic) lemma, which completes the proof.

2185: \QED

2186: 

2187: \begin {lemma}\label{l.deterministic}

2188: Let $D_1\supseteq D_2 \supseteq D_3 \ldots$

2189: be a decreasing

2190: sequence of compact subsets of $[0,1]$,

2191: and let $\mu_1,\mu_2,\dots$ be a sequence of positive measures

2192: with $\mu_n$ supported on $D_n$.

2193: Suppose that there is a constant $C$ such that for infinitely many

2194: values of $n$, we have

2195: \begin{equation}\label{e.2cond}

2196: \| \mu_n \| > 1/C,\quad \text{ and } \quad{\cal E}_{\gamma} (\mu_n)\le C.

2197: \end{equation}

2198: Then the Hausdorff dimension of $\bigcap_n D_n$ is at least $\gamma$. 

2199: \end {lemma}

2200: 

2201: \proof

2202: Choose a sequence of integers $\{n_k\}$ for which

2203: (\ref{e.2cond}) holds.

2204: Note that $\|\mu_{n_k}\|^2\le \mathcal E_\gamma(\mu_{n_k})\le C$.

2205:  By compactness, 

2206: choose a further subsequence $\{n'_{k}\}$ of $\{n_k\}$ so that

2207: $\mu_{n'_{k}}$ converges weakly to some positive measure $\mu_{\infty}$.

2208: Clearly $\mu_{\infty}$ is supported on $\bigcap_n D_n$ and

2209: $\| \mu_\infty \| \ge 1/C$. For all $M$, we have that

2210: \begin{multline*}

2211: \int \!\!\int {|x-y|}^{- \gamma} \wedge M \, d\mu_\infty (x) \,   d\mu_\infty(y) =

2212: \\ \lim_{k\to\infty}

2213: \int \!\!\int {|x-y|}^{- \gamma} \wedge M \, d\mu_{n'_k} (x) \,   d\mu_{n'_k}(y) \le C.

2214: \end{multline*}

2215: Now let $M\to\infty$ and apply the monotone convergence theorem to conclude that

2216: $$

2217: \int \!\!\int {|x-y|}^{- \gamma}\, d\mu_\infty (x) \,   d\mu_\infty(y) 

2218: \le C.

2219: $$

2220: Since $\| \mu_\infty \| > 0$, it now follows from Frostman's theorem

2221: (see for example,~\cite{Kahane})

2222: %(see for example, \cite{Kahane}, page 133)

2223: that the Hausdorff dimension of $\bigcap_n D_n$ is at least $\gamma$. 

2224: \QED

2225: 

2226: \begin{theorem}\label{t.upperbound}

2227: A.s., 

2228: the Hausdorff dimension of the set of exceptional times is at most $31/36$.

2229: \end{theorem}

2230: 

2231: \proof

2232: Let $U_n$ be the event that there is a time in $[0,1/n]$ for which

2233: the origin percolates. Since the set of vertices which are open for some $t\in [0,1/n]$

2234: is an i.i.d.\ process with density $1/2 +(1-e^{-1/(2n)})/2\le 1/2+1/n$, it is immedate that

2235: $$

2236: \Pb{U_n}\le \pi_{\frac{1}{2}+\frac{1}{n}} (\calC_0) ,

2237: $$

2238: where $\calC_0$ is the event that the origin percolates.

2239: By page 3 of \cite{SW}, for every $\epsilon >0$, there is a $C$ so that

2240: \begin{equation}\label{e.offcrit}

2241: \pi_{\frac{1}{2}+\frac{1}{n}} (\calC_0)\le 

2242: C\, n^{\eps-5/36}.

2243: \end{equation}

2244: Now let 

2245: $$

2246: N_n=\sum_{j=1}^n 1_{U_{j,n}}\,,

2247: $$

2248: where $U_{j,n}$ is the event that there is a time in $[(j-1)/n,j/n]$ for which

2249: the origin percolates (so $U_{1,n}=U_n$ above). 

2250: By the above, we have that $\Eb{N_n}\le C n^{\frac{31}{36}+\epsilon}$.

2251: It follows that

2252: $$

2253: \lim_n \frac{\Eb{N_n}}{n^{\frac{31}{36}+2\epsilon}}=0

2254: $$

2255: and so from Fatou's lemma, we get

2256: $$

2257: \EB{\liminf_n \frac{N_n}{n^{\frac{31}{36}+2\epsilon}}}=0.

2258: $$

2259: Therefore

2260: $$

2261: \liminf_n \frac{N_n}{n^{\frac{31}{36}+2\epsilon}}=0

2262: $$

2263: a.s. This says that a.s.\ for infinitely many $n$, the set of exceptional

2264: times in $[0,1]$ at which the origin percolates

2265: can be covered by $n^{\frac{31}{36}+2\epsilon}$ intervals of length $1/n$.

2266: Hence, the Hausdorff dimension of the set of these exceptional times is 

2267: at most $\frac{31}{36}+2\epsilon$ a.s. By countable additivity, we are done.

2268: \QED

2269: 

2270: \begin{remark}

2271: The upper bound will be proved again by a different argument when

2272: we prove Theorem~\eref{th:cones}.

2273: The above proof is included 

2274: here, because it is shorter.

2275:  One should nonetheless

2276: point out that the above argument uses~\eref{e.offcrit},

2277: while the argument below is more self contained.

2278: \end{remark}

2279: 

2280: \section{Exceptional times for $k$-arm events} \label{sec:wedgeslower}

2281: 

2282: 

2283: In this section, we give the proofs of the lower bounds in

2284: Theorems \ref{th:wedges} and \ref{th:cones},

2285: but generally omit those details which are the same as in the

2286: corresponding proofs of Theorems~\ref{th:except} and~\ref{t.lowerbound}.

2287: 

2288: For $\theta>0$ and integer $k\ge 1$, let

2289: $A_{W_\theta}^k(r,R)$ be the event that

2290: we have $k$ disjoint crossings of alternating colors (with black most clockwise)

2291: between distances $r$ and $R$ of the origin in $W_\theta$

2292: and let $\alpha_{W_\theta}^k(r,R)=\Pb{A_{W_\theta}^k(r,R)}$. 

2293: If $r$ is suppressed, then it is taken to be $10\,k$.

2294: 

2295: We will, of course, need the asymptotics of $\alpha_{W_\theta}^k(r,R)$.

2296: For this purpose, conformal invariance will be used.

2297: Although when $\theta>2\pi$ the surface

2298: $W_\theta$ is not planar and conformal invariance is usually stated for planar domains,

2299: the proof of conformal invariance certainly holds in this setting.

2300: The asymptotic decay as $R/r\to\infty$ of the probability of $k$ disjoint crossings

2301: between distances $r$ and $R$ in $W_{\theta}$ from the origin in 

2302: the percolation scaling limit is determined using conformal invariance.

2303: Specifically, the map $z\mapsto z^{\pi/\theta}$ maps $W_{\theta}$ to the

2304: upper half plane, and we may conclude from~\eref{e.halfpk}

2305: that the decay (for the percolation

2306: scaling limit)  is of the form $(R/r)^{\frac{-\pi k(k+1)}{6\theta}+o(1)}$,

2307: as $R/r\to\infty$ while $k$ stays fixed.

2308: Then, one can conclude, as for the other exponents we have discussed,

2309: that for $R\ge r\ge 10\,k$,

2310: \begin{equation}\label{e.awt}

2311: \alpha_{W_\theta}^k(r,R)= (R/r)^{\frac{-\pi k(k+1)}{6\,\theta}+o(1)},

2312: \end{equation}

2313: as $R/r\to\infty$ while $k$ is fixed, using the argument given in~\cite{SW}. 

2314: We will also use the fact that the quasi-multiplicativity relations~\eref{e.multip}

2315: and~\eref{e.rsw} hold for $\alpha_{W_\theta}^k$ and

2316: for $\alpha_2$. This is proved in the appendix;

2317: see Remark~\ref{r.qmc}.

2318: 

2319: 

2320: \proofof{lower bound in Theorem~\ref{th:wedges}}

2321: We first handle the case $k=1$ and therefore abbreviate temporarily

2322: $\alpha_{W_\theta}^1(r,R)$ by $\alpha_{W_\theta}(r,R)$.

2323: (A different approach will be needed for $k\ge 2$.)

2324: Let $D$ be the union of the hexagons in $W_\theta$ that contain

2325: points whose distance from the origin is in $[r,R]$.

2326: Let $\p_R D$ and $\p _r D$ denote the set of points in

2327: $\p D$ that are at distance $\ge R$ [respectively,

2328: $\le r$] from the origin.

2329: Also, we denote by $\p^0 D$ and $\p^\theta D$,

2330: the components of $\p D\cap\p W_\theta$ that are at angle

2331: about $0$ [respectively, about $\theta$] in radial coordinates on $W_\theta$.

2332: 

2333: The algorithm we use to determine if

2334: there exists a crossing of $D$ is essentially the same as

2335: the algorithm determining the existence of a left to right crossing of a square,

2336: where $\p_R D$

2337: plays the role of the right side of the square and

2338: where $\p_r D$

2339: plays the role of the left side of the square.

2340: (This is of course crucial; if we reversed things, then the hexagons

2341: near the inner circle would be revealed with too high of a probability.)

2342: It is clear that this algorithm works and so we now need

2343: to compute its revealment.

2344: We will show that the revealment is

2345: \begin{equation}\label{e.revealment}

2346: O(1)\,\alpha_2(r)\,\alpha_{W_\theta}(r,R)\,.

2347: \end{equation}

2348: Using~\eref{e.awt} and~\eref{e.expostrong}, one can show that this is

2349: essentially (i.e., up to some $O(1)$ factor) monotone decreasing

2350: in $r$ in the relevant range $\theta>8\,\pi/3$.

2351: 

2352: We just look at the first interface arising in the algorithm,

2353: the one which terminates when it hits $\p_r D\cup\p^\theta D$,

2354: since the estimates for the second interface will be essentially the same.

2355: 

2356: Fix some hexagon $H\subset D$.

2357: Let $s=\dist(H,\p_R D\cup \p^0 D\cup \{0\})$ with $0$ denoting the origin.

2358: We also use $|p|$ to denote distance from $0$ in $W_\theta$.

2359: We distinguish several different cases.

2360: 

2361: \medskip\noindent

2362: {\bf Case 1}: $\dist(H,\{0\})=s$. 

2363: For $H$ to be visited, we need our 2 arms event holding within distance $s/2$ of $H$

2364: and a crossing of the desired color between distance $2s$ and distance $R$ from the 

2365: origin. These are independent and we get that $H$ is visited with probability 

2366: at most $\alpha_2(s/2)\,\alpha_{W_\theta}(2s,R)$.

2367: By the analogues of~\eref{e.multip} and~\eref{e.rsw} for

2368: $\alpha_2$ and $\alpha_{W_\theta}$, this is compatible with

2369: our claimed revealment~\eref{e.revealment}.

2370: 

2371: 

2372: \medskip\noindent

2373: {\bf Case 2}: $\dist(H,\p_R D)=s$. 

2374: As in the proof of Theorem~\ref{t.annulus} with $c:=\dist(p_0,H)\wedge (R/2)$,

2375: we obtain

2376: $$

2377: \Pb{H\text{ visited }\md p_0}\le O(1)\,\alpha^+(2s+O(1),c-s-O(1))\,\alpha_2(s/2).

2378: $$

2379: Proceeding as in that proof, this is also compatible with

2380: our claimed revealment~\eref{e.revealment}.

2381: 

2382: \medskip\noindent

2383: {\bf Case 3}: $\dist(H,\p^0 D)=s$. 

2384: Let $w\in \p^0 D$ so that $\dist(H,w)=s$. We separate Case 3 into 3 subcases.

2385: 

2386: \medskip\noindent

2387: {Case 3(a)}: $s \ge |w|/2$.

2388: Then the triangle inequality gives $\dist(H,0)\le 3s$. 

2389: For $H$ to be visited, we need our 2 arms event holding within distance $s/2$ of 

2390: $H$ and a path of the desired type between distance $4s$ and distance $R$ 

2391: from the origin. These are independent and we get that $H$ is visited with 

2392: probability at most $\alpha_2(s/2)\,\alpha_{W_\theta}(3s,R)$,

2393: which is compatible with our claimed revealment~\ref{e.revealment}.

2394: 

2395: \medskip\noindent

2396: {Case 3(b)}: $s \le |w|/2 \le R/4$.

2397: For $H$ to be visited, we need our 2 arms event holding within distance $s/2$ of $H$,

2398: a white crossing in the half annulus centered at $w$ with outer radius $|w|$ and inner radius $2s$

2399: (which is identical to a half-annulus in a half-plane;

2400: $|w|\le R/2$ guarantees that the above half-annulus does not intersect $\p_R D$)

2401: and a white crossing between distance $2|w|$ and distance $R$ from the 

2402: origin.

2403: These are independent and we

2404: get that $H$ is visited with probability at most 

2405: $$

2406: \alpha_2(s/2)\,\alpha_{+}(2s,|w|)\, \alpha_{W_\theta}(2|w|,R).

2407: $$

2408: Since up to an $O(1)$ factor $\alpha_2(s)\,\alpha_{+}(s,|w|)$ is increasing in $s$,

2409: the product of the first two terms is at most

2410: $O(1)\,\alpha_2(|w|)$ and since $|w|\ge 2\,s\ge r$, the whole product is at most 

2411: $$

2412: O(1)\,\alpha_2(r)\,\alpha_{W_\theta}(r,R).

2413: $$

2414: 

2415: \medskip\noindent

2416: {Case 3(c)}: $|w| \ge R/2$; $s\le |w|/2$.

2417: For $H$ to be visited, we need our 2 arms event holding 

2418: within distance $s/2$ of $H$ and if 

2419: $2s<d(p_0,w)$ it is also necessary that

2420: a white crossing 

2421: occurs between distances $2s$ and $d(p_0,w)\wedge |w|$ from

2422: $w$.

2423: (Note that the latter event takes place in the upper half plane.)

2424: These are independent and since $|w|\ge R/2$ we get 

2425: $$

2426: \Pb{H\text{ visited }\md p_0}\le O(1)\,\alpha^+\bl(2s,d(p_0,w)\wedge (R/2)\br)\,\alpha_2(s/2)\,.

2427: $$

2428: As in Case 2, this is compatible with~\eref{e.revealment}.

2429: 

2430: \medskip

2431: This covers all possible cases, and hence establishes that the

2432: revealment is as claimed.

2433: 

2434: \medskip

2435: 

2436: We now proceed to discuss the algorithm and the revealment when $k>1$.

2437: It turns out simplest in fact to

2438: modify the event $A_{W_\theta}^k(r,R)$ as follows. Partition  the

2439: outer boundary $\p_R D$ into $k$ arcs  of roughly equal diameter

2440: $Y_1,Y_2,\ldots,Y_k$ (ordered counterclockwise) and let $\tilde{A}_{W_\theta}^k(r,R)$

2441: be the event that for every odd [respectively, even] $i\in\{1,2,\dots,k\}$ there is a black

2442: [respectively, white]

2443: crossing in $D$ from $\p_r D$ to $Y_i$.

2444: Thus, instead of looking at the set of times for which $A_{W_\theta}^k(r_0,R)$

2445: occurs (where $r_0=10\,k$, say), we will look at the set of times at which $\tilde A_{W_\theta}^k(r_0,R)$

2446: occurs.

2447: Clearly, $\tilde{A}_{W_\theta}^k(r,R)\subset A_{W_\theta}^k(r,R)$,

2448: and therefore this is justified.

2449: We will also use the relation

2450: \begin{equation}\label{e.AA}

2451: \Pb{{A}_{W_\theta}^k(r,R)} \le C^\theta_k\, \Pb{ \tilde A_{W_\theta}^k(r,R)}\,,

2452: \end{equation}

2453: for some constant $C_k^\theta$, depending only on $k$ and $\theta$,

2454: which holds by Remark~\ref{r.primus}.

2455: 

2456: If $Y\subset \p_R D$ is an arc, let $A_Y^1(r,R)$

2457: [respectively, $A_Y^{-1}(r,R)$] be the

2458: event that there is a white [repectively, black] crossing from $Y$ to $\p_r D$

2459: in $D$.

2460: Suppose that each for each $i=1,2,\dots,k$, we have

2461:  a partition $Y_i=Y_{i,+}\cup Y_{i,-}$

2462: of $Y_i$ into two arcs $Y_{i,+}$ and $Y_{i,-}$.

2463: Since $A_{Y_i}^{\pm 1}(r,R)=A_{Y_{i,+}}^{\pm 1}(r,R)\cup A_{Y_{i,-}}^{\pm 1}(r,R)$,

2464: we have

2465: \begin{equation}\label{e.union}

2466: \tilde{A}_{W_\theta}^k(r,R)

2467: =\bigcup_{y\in \{-,+\}^k}

2468: \bigcap_{i=1}^k A_{Y_{i,y_i}}^{{(-1)}^i}(r,R)\,.

2469: \end{equation}

2470: 

2471: The algorithm starts out by picking points $x_i\in Y_i$,

2472: randomly, uniformly and independently.

2473: Then $Y_{i,+}$ and $Y_{i,-}$ are chosen as the two components

2474: of $Y_i\setminus \{x_i\}$.

2475: For each of the $2^k$ possible $y\in\{-,+\}^k$,

2476: the algorithm then proceeds to determine if

2477: the corresponding component

2478: $$

2479: A(y):=

2480: \bigcap_{i=1}^k A_{Y_{i,y_i}}^{{(-1)}^i}(r,R)

2481: $$

2482: of~\eref{e.union} has occured.

2483: For that purpose, interfaces are started at each of the points $x_i$,

2484: and are followed until the event has been determined one way or the other.

2485: (Of course, the interface will have either white on the left and

2486: black on the right or vice versa, depending on the color of crossing

2487: it is meant to detect and whether the corresponding arc

2488: $Y_{i,\pm}$ is to the left or right of $x_i$.)

2489: However, the order in which the interfaces are extended is somewhat important.

2490: A simple rule that works is that among the hexagons that are necessary

2491: to extend the $k$ interfaces one more step, the algorithm chooses the one that is

2492: farthest away from $0$.

2493: The event $A(y)$ is decided positively only if

2494: all $k$ interfaces reach $\p_r D$.

2495: 

2496: The revealment of this algorithm is at most $k\,2^k$ times the maximum

2497: probability that the interface started at $x_i$ visits a hexagon

2498: $H$ before the determination of the corresponding

2499: $A(y)$ is terminated. Here, the maximum is over all hexagons $H\subset D$

2500: and all $i\in\{1,2,\dots,k\}$.

2501: The corresponding bound is attained as in the case

2502: $k=1$, but now $\alpha_{W_\theta}^1$ is replaced by

2503: $\alpha_{W_\theta}^k$. 

2504: Our rule of thumb for selecting which interface to

2505: extend guarantees that we never examine a hexagon

2506: $H$ unless $\tilde A_{W_\theta}^k\bl(\dist(0,H)+O(1),R\br)$ occurred.

2507: As in the case $k=1$, when estimating the revealment it is important that

2508: $\alpha_2(r,R)\le O(1)\,\alpha_{W_\theta}^k(r,R)$.

2509: In the range $\theta> 4\,\pi\,k\,(k+1)/3$, which is the

2510: relevant range for the lower bound in Theorem~\ref{th:wedges}, this follows

2511: from~\eref{e.awt}.

2512: 

2513: The remainder of the proof goes through as before.

2514: \QED

2515: 

2516: \proofof{lower bound in Theorem~\ref{th:cones}}

2517: Here we simply say that the proof for the lower bounds in 

2518: Theorem~\ref{th:wedges} can be carried out in a similar way. 

2519: In fact, for $k\ge 2$, the proof

2520: is simpler topologically than the $k=1$ case for the plane,

2521: since we do not need to worry about interfaces making complete circuits

2522: around the origin 

2523: (if this ever happens, the event in question cannot occur and we stop the 

2524: algorithm).

2525: \QED

2526: 

2527: \section{Upper bounds for $k$-arm times} \label{sec:wedgesupper}

2528: 

2529: The following result,

2530: which will be useful for the proofs

2531: of the upper bounds in Theorems~\ref{th:wedges} and~\ref{th:cones},

2532: is abstract: the graph structure does not play any role.

2533: Let $A$ be an event involving independent Bernoulli $(1/2,1/2)$ random variables

2534: $X_1,X_2,\ldots,X_m$. Recall that the influence of the index $i$ on $A$, denoted $I_i(A)$,

2535: is the probability that $X_i$ is pivotal; namely,

2536: that changing the value of $X_i$ changes whether $A$ 

2537: occurs or not. 

2538: The sum of the influences is denoted by $I(A)=\sum_i I_i(A)$.

2539: 

2540: 

2541: 

2542: \begin{theorem} \label{th.hdgeneral}

2543: Let $\{A_n\}_{n\ge 1}$ be some sequence of events 

2544: in $\{0,1\}^V$, each depending on only finitely many 

2545: coordinates. 

2546: Assume that $\lim_{n\to\infty} \Ps{A_n} = 0$.

2547: Let $\omega_t$ be the Markov process on $\{0,1\}^V$ where

2548: independently $0$'s go to $1$ at rate $1/2$,

2549: $1$'s go to $0$ at rate $1/2$ and started according

2550: to its stationary distribution $\pi_{\frac{1}{2}}$.

2551: Let $T$ be the set of exceptional times $t$

2552: at which $\omega_t\in\bigcap_{n\ge 1}A_n$.

2553: If $\liminf_{n\to\infty} I(A_n) < \infty$, then 

2554: $T=\emptyset$ a.s. Otherwise, the Hausdorff dimension of $T$ is a.s.\ at most

2555: \begin{equation}\label{e.expression}

2556: \liminf_{n\to\infty}

2557: \left({{1-\frac{\log \Ps{A_n}}{\log I(A_n)}}}\right)^{-1}\,.

2558: \end{equation}

2559: \end{theorem} 

2560: 

2561: 

2562: \proof

2563: Let $T_n:=\{t\in [0,1]:\omega_t\in A_n\}$, let $\partial T_n$ be the boundary

2564: points of $T_n$ in $(0,1)$ and set $N_n:=|\partial T_n|$. We claim that

2565: \begin{equation}\label{e.Nn=In} 

2566: \Eb{N_n}={I(A_n)}/{2}.

2567: \end{equation}

2568: To see this, write

2569: $N_n=\sum_v N^v_n$ where $N^v_n$ counts the number of elements in

2570: $\partial T_n$ at which time the vertex $v$ flipped. We now need to

2571: show that, for each vertex $v$, $\Eb{N^v_n}$ is $I_v(A_n)/2$.

2572: Given a time interval $[t,t+dt]$, the probability that 

2573: there is a time point in the interval which contributes to $N^v_n$

2574: is equal to $I_v(A_n)\,dt/2 + o(dt)$ and the probability of $k\ge 2$ such time

2575: points is clearly $O(dt^k)$. From this,~(\ref{e.Nn=In}) easily follows.

2576: 

2577: For any $\eps > 0$, let $T^\eps_n$ be the $\eps$-neighborhood of

2578: $T_n$ intersected with $[0,1]$.

2579: Since $T^\eps_n\subseteq T_n\cup \bigcup_{x\in \p T_n}[x-\eps,x+\eps]$,

2580: \begin{equation}\label{e.star} 

2581: \mu(T^\eps_n)\le \mu(T_n)+ 2\,N_n\,\eps\,,

2582: \end{equation}

2583: where $\mu$ denotes Lebesgue measure.

2584: For any set $U$ and $\epsilon>0$, let $\calN(U,\epsilon)$ denote the number of 

2585: $\epsilon$ intervals needed to cover $U$.

2586: From the above, using the fact that the intervals comprising

2587: $T^\eps_n$ all have length at least $\eps$, it follows that

2588: $\calN(T^\eps_n,\eps)\le2\,\mu(T^\eps_n)\,\eps^{-1}$, and so,

2589: using~\ref{e.star},

2590: $$

2591: \calN(T_n,\eps)\le \calN(T^\eps_n,\eps)\le

2592: {2\,\mu(T_n)\,}{\eps^{-1}}+ 4\,N_n\,.

2593: $$

2594: Therefore, by Fubini's theorem and~\ref{e.Nn=In}, 

2595: \begin{equation}\label{e.covernumber}

2596: \Eb{\calN(T_n,\eps)}\le {2\,\Ps{A_n}}\,{\eps^{-1}}+ 2\,I(A_n)\,.

2597: \end{equation}

2598: 

2599: We now temporarily assume that $\liminf_{n\to\infty} I(A_n) =\infty$. 

2600: Let $a_n=\Pb{A_n}/I(A_n)$, which goes to $0$ as $n\to\infty$.

2601: By (\ref{e.covernumber}), we have 

2602: \begin{equation}\label{e.neededbound}

2603:   \Eb{\calN(T,a_n)}\le 

2604:   \Eb{\calN(T_n,a_n)}\le 4\,I(A_n)\,.

2605: \end{equation}

2606: By passing to a subsequence if necessary, we assume with no loss of generality

2607: that the $\liminf$ in~\eref{e.expression} is a limit. Let $L$ denote the value

2608: of that limit.

2609: It is elementary to check that for every $\eps>0$, for all sufficiently large $n$,

2610: $$

2611: I(A_{n})\le \left(\frac{I(A_{n})}{\Ps{A_{n}}}\right)^{L+\epsilon}.

2612: $$

2613: This together with (\ref{e.neededbound})

2614: implies that  %

2615: the Hausdorff dimension of $T$ is at most $L+\epsilon$ a.s. As 

2616: $\epsilon$ is arbitrary, this completes the proof in the case

2617: $I(A_n)\to\infty$.

2618: 

2619: Since $T_n\ne \emptyset$ implies that $N_n\ge 1$ or $T_n\supseteq (0,1)$,

2620: it follows by (\ref{e.Nn=In}) and Markov's inequality that 

2621: $$

2622: \Pb{T_{n} \neq \emptyset}\le \Pb{A_{n}} + I(A_{n})\,.

2623: $$

2624: Thus, $T=\emptyset$ a.s.\ when $\liminf_n I(A_n)= 0$.

2625: 

2626: The case $\liminf_n I(A_n)\in (0,\infty)$ requires a different argument.

2627: Let $\eps_n=\sqrt{\Ps{A_n}}$. By (\ref{e.covernumber}), we have 

2628: $

2629: \liminf_{n\to \infty}\Eb{\calN(T_n,\eps_n)} < \infty

2630: $.

2631: Since $\eps_n\to 0$, the cardinality $|T|$ of $T$ is bounded from above by

2632: $ \liminf_{n\to\infty}\calN(T_n,\eps_n) $.

2633: Fatou's lemma yields $\Eb{|T|}< \infty$ and

2634: hence $\Pb{|T|<\infty}=1$. We finally conclude that

2635: $\Ps{T\neq\emptyset}=0$ by combining

2636: \cite[Theorem 6.7]{GS} and \cite[(2.9)]{FG}.

2637: \QED

2638: 

2639: \proofof{Theorem~\ref{th:cones}}

2640: Since the lower bounds have been established in Section~\ref{sec:wedgeslower},

2641: it remains to prove the upper bounds.

2642: Fix $k=1$ or $k>1$ even.

2643: Let $A_R$ be the event that there are $k$ disjoint

2644: crossings of the annulus $D_R:=\{z\in C_\theta: 10k\le |z|\le R\}$, where we require

2645: that the colors be alternating if $k\ne 1$.

2646: Here, $|z|$ denotes the distance to $0$, which is the apex of the cone $C_\theta$.

2647: One can prove that 

2648: \begin{equation}\label{e.coneexpo}

2649: \Pb{A_R}

2650: =\begin{cases}

2651: R^{-5\pi/(24\theta)+o(1)}&k=1,\\

2652: R^{{(1-k^2)\pi}/({6\theta}) + o(1)}& k>1,

2653: \end{cases}

2654: \end{equation}

2655: in the very same way that we have justified~\eref{e.awt}.

2656: By Theorem~\ref{th.hdgeneral} (and easy algebraic manipulation), it therefore suffices to prove

2657: that

2658: \begin{equation}\label{e.pivotal}

2659: I(A_R)/\Ps{A_R}\le R^{3/4+o(1)}.

2660: \end{equation}

2661: Let $H$ be a hexagon in $C_\theta$, and let $s=s(H)$ be the distance from

2662: $H$ to $0$.

2663: For $H$ to be pivotal 

2664: it is necessary that there would be

2665: $k$ disjoint (alternating, if $k>1$)

2666: crossings from distance $10\,k$ to

2667: $s/2$ from the origin (unless $s/2\le 10\,k$)

2668: and between distances $2\,s$ and

2669: $R$ (unless $2\,s\ge R$).

2670: Likewise, there should be $4$ alternating crossings

2671: between $H$ and distance

2672: $(s/2)\wedge\dist(H,\p D_R)$ from $H$.

2673: These events are independent.

2674: Using the quasi-multiplicative property

2675: of the $k$-arm crossing events (Remark~\ref{r.qmc})

2676: and~\eref{e.fullpk} with $k=4$,

2677: this gives when $s<8\,R/9$

2678: \begin{equation}\label{e.sumI}

2679: I_H(A_R)\le O(1)\,\Ps{A_R}\,s^{-5/4+o(1)}.

2680: \end{equation}

2681: where this $O(1)$ factor (as well as those appearing

2682: below) may depend on $k$ and $\theta$.

2683: Since the number of hexagons in $C_\theta$

2684: satisfying $s=s(H)<\rho$ is $O(\rho^2)$,

2685: an easy calculation yields

2686: $$

2687: \sum_{H:s(H)< 8R/9} I_H(A_R)\le O(1)\,\Ps{A_R}\,R^{3/4+o(1)}\,.

2688: $$

2689: 

2690: Now suppose that $H$ is a hexagon satisfying

2691: $s(H)\ge 8\,R/9$.

2692: For $H$ to be pivotal, it is necessary that there would

2693: be $k$ (alternating, if $k>1$) crossings in $C_\theta$

2694: between $\{|z|=10\,k\}$ and $\{|z|=R/2\}$,

2695: there should be $4$ alternating crossings

2696: between $H$ and distance $\dist(H,\p D_R)/2$ from $H$,

2697: and there should be $3$ alternating

2698: crossings

2699: between distance $2\,\dist(H,\p D_R)$ and

2700: distance $R/2$ from a point on $\p D_R$ closest to $H$.

2701: The latter event is governed by the $3$-arm half

2702: plane exponent, whose asymptotic behaviour is described by~\eref{e.halfpk}.

2703: Since $s+\dist(H,\p D_R)=R+O(1)$, we get

2704: $$

2705: I_H(A_R)\le O(1)\,\Ps{A_R}\,(R-s)^{-5/4+o(1)}\,((R-s)/R)^{2+o(1)}\,.

2706: $$

2707: Since for $b\ge 1$ there are $O(b\,R)$ hexagons at distance $\le b$ from

2708: $\{|z|=R\}$, another easy calculation gives

2709: $$

2710: \sum_{H:s(H)\ge  8R/9} I_H(A_R)\le O(1)\,\Ps{A_R}\,R^{3/4+o(1)}\,.

2711: $$

2712: Together, this yields~\eref{e.pivotal} and the proof is complete.

2713: \QED

2714: 

2715: 

2716: \proofof{Theorem~\ref{th:wedges}}

2717: The lower bound was proved in Section~\ref{sec:wedgeslower},

2718: and so only the upper bound needs to be justified.

2719: The proof proceeds like the proof of the upper bound in Theorem~\ref{th:cones},

2720: except that the influence estimates are slightly different.

2721: 

2722: Let $D_R=\{z\in W_\theta:10\,k\le |z|\le R\}$,

2723: $A_R$ be the $k$-arm event in $W_\theta$

2724: between $\{z:|z|=10\,k\}$ and

2725: $\{z:|z|=R\}$, and $H\subset D_R$ be a hexagon.

2726: Let $s=s(H)=\dist (0,H)$, and let $b=b(H)=\dist (H,\p D_R)$,

2727: where we write $\p D_R$ for the boundary of $D_R$ in $C_\infty$,

2728: i.e., the points on $\p W_\theta$ are included.

2729: For $H$ to be pivotal for $A_R$, it is necessary

2730: that the $k$ arm event holds

2731: between distance $10\,k$ and $s/2$ from $0$ (unless

2732: $s/2\le 10\,k$) as well as between

2733: distances $2\,s$ and $R$ (unless $2\,s\ge R$),

2734: that the alternating $4$-arm

2735: event hold between $H$ and distance $b/2$ away from $H$,

2736: and that the alternating $3$-arm event must hold

2737: between distances $2\,b$ and $s/4$ away from a point

2738: in $\p D_R$ closest to $H$ (unless $2\,b\ge s/4$).

2739: There are $O(b's')$ hexagons $H$ satisfying $b(H)\le b'$

2740: and $s(H)\le s'$.

2741: The rest of the proof proceeds like that of Theorem~\ref{th:cones},

2742: and is left to the reader.

2743: \QED

2744: 

2745: 

2746: \proofof{Theorems~\ref{th.no2clusters},

2747: \ref{th.differentclusters},

2748: \ref{th.halfplane} and

2749: \ref{th.halfplane2}}

2750: At any time at which there are 2 infinite white clusters in the plane,

2751:  we must also have

2752: the 4-arm event occuring (with alternating colors) but by

2753: Theorem~\ref{th:cones} (with $k=4$ and $\theta=2\pi$), there are no such times.

2754: This proves Theorem~\ref{th.no2clusters}.

2755: 

2756: At any time at which there are 2 infinite different colored clusters, we must also 

2757: have the 2-arm event occuring (with different colors) but by

2758: Theorem \ref{th:cones} (with $k=2$ and $\theta=2\pi$), the set of

2759: such times has Hausdorff dimension at most $2/3$.

2760: This proves Theorem~\ref{th.differentclusters}.

2761: The other two theorems are similarly proved.

2762: \QED

2763: 

2764: 

2765: 

2766: 

2767: \section{The square lattice} \label{sec:square}

2768: 

2769: We start this section by proving 

2770: Theorem~\ref{th.no3clustersZ2}.

2771: Afterwards, possible ways in which our arguments for

2772: Theorem~\ref{th:except} may be improved to apply to $\Z^2$ 

2773: as well, will be discussed.

2774: 

2775: In the proof of Theorem~\ref{th.no3clustersZ2} we will

2776: use the fact that the $6$ alternating arms exponent is

2777: larger than $2$, or, more precisely, 

2778: that the probability for $6$ alternating arms

2779: between radii $r$ and $R$ is

2780: bounded above by $O(1)\,(r/R)^{2+\eps}$ for some $\eps >0$.

2781: This is essentially due to~\cite[Lemma 5]{KSZ},

2782: but a proof is also given in the appendix (Corollary~\ref{c.56}).

2783: 

2784: 

2785: \proofof{Theorem~\ref{th.no3clustersZ2}}

2786: For $0<r<R$, let $S(r,R)$ be the event

2787: that there are $3$ different clusters that connect the circles

2788: of radii $r$ and $R$ about $0$.

2789: By the above mentioned bound on the alternating $6$-arm probabilities,

2790: We may choose some fixed $\eps>0$ and

2791: some function $\rho=\rho(r)>r$ such that for static critical bond percolation on $\Z^2$,

2792: for all $r$,

2793: \begin{equation}\label{e.rhochoice}

2794: \Pb{ S(r,\rho)}\le\rho^{-2-\eps}\,.

2795: \end{equation}

2796: Consider some bond $e$, and let $F(e)$ be the event that

2797: $e$ is pivotal for $S(r,\rho)$.

2798: Then $\Pb{F(e)}$ is just the influence $I_e(S(r,\rho))$.

2799: Assume that $\Pb{F(e)}\ne 0$.

2800: Note that the events $F(e)$ and $\{e\text{ is open}\}$ are independent

2801: events. This implies that $\Pb{S(r,\rho)\md F(e)}=1/2$,

2802: which one may write

2803: $ \Pb{F(e)\cap S(r,\rho)}=\Pb{F(e)\cap\neg S(r,\rho)}$.

2804: Since this applies to every bond $e$, we conclude that

2805: the expected number of pivotals on the event $S(r,\rho)$

2806: is half of the total influence $I(S(r,\rho))$.

2807: However the number of pivotals for $S(r,\rho)$ is bounded

2808: by the total number of edges intersecting the

2809: disk of radius $\rho$ about the origin, which is

2810: certainly $O(\rho^2)$. Thus, 

2811: $$

2812: I(S(r,\rho))\le 2\,\Pb{S(r,\rho)}\,O(\rho^2) =O(\rho^{-\eps}).

2813: $$

2814: Consequently, by Theorem~\ref{th.hdgeneral}, for every $r_0>0$ a.s.\ there

2815: are no exceptional times in which

2816: $\bigcap_{r>r_0} S(r,\rho(r))$ holds.

2817: This proves our theorem.

2818: \QED

2819: 

2820: 

2821: 

2822: \begin{remark}

2823: An alternative way to prove the above result is based on using the fact that the 6-arm 

2824: exponent is strictly larger than 2 together with the fact that the number of different

2825: configurations (counting repetitions) that appear in a ball of radius 

2826: $n$ during the time interval $[0,1]$ has a Poisson distribution

2827: with a parameter which is at most $O(1) n^2$.

2828: \end{remark}

2829: 

2830: 

2831: \bigskip

2832: 

2833: As we will briefly explain below, the proof of

2834: Theorem~\ref{th:except} almost works for bond percolation on the square

2835: grid. In fact, there are several alternative routes by which

2836: the result might perhaps be extended to $\Z^2$:

2837: \begin{enumerate}

2838: \item establishing 

2839: \begin{equation}\label{e.exponeed}

2840: \alpha_2(r)\le O(r^{-\eps})\,\alpha(r)^{2}

2841: \end{equation}

2842:  for $\Z^2$ for some fixed $\eps >0$,

2843: \item improving the estimate~\eref{e.noise},

2844: \item proving the existence of an algorithm

2845: (or a witness which would still permit the use of

2846: Theorem~\ref{t.noise}) with smaller revealment,

2847: \item extending Smirnov's theorem to $\Z^2$.

2848: \end{enumerate}

2849: 

2850: Note that the weaker version of~\eref{e.exponeed}

2851: $\alpha_2(r)\le \alpha(r)^2$ follows from either the Harris-FKG 

2852: inequality or Reimer's inequality~\cite{Reimer}.

2853: Kesten and Zhang have proved some related strict inequalities

2854: between exponents~\cite{KZ}, but it seems that their

2855: methods are not sufficient to prove~\eref{e.exponeed}.

2856: 

2857: We now explain why~\eref{e.exponeed} in the $\Z^2$ setting implies

2858: exceptional times for $\Z^2$.

2859: First we want to have the revealment

2860: for the algorithm determining $\croRr$ bounded by

2861: $O(1)\,\alpha_2(r)\,\alpha(r,R)$.

2862: One problem seems to be that

2863: the bound on the revealment

2864: for the triangular grid involves the

2865: summand featuring $\alpha^+$, which is relatively negligible, while

2866: on $\Z^2$, we do not know how to prove that the other summand

2867: dominates. The fix is to replace the deterministic $R$

2868: by a random $R_t'\in[R,2R]$.

2869: The random variable $R_t'$ will depend on some extra random bits,

2870: that we add, and these random bits also evolve in time.

2871: We construct the dependence of $R'$ on these bits

2872: so that $R'$ can be calculated by an algorithm with very

2873: small revealment.

2874: This is rather easy to arrange, because we are not limited 

2875: in the number of bits that we may take. 

2876: If we consider an edge whose distance from the origin $a$ is

2877: in the range $[R/2,2R]$, then the probability

2878: that the edge is examined given $R'$ is at most

2879: $O(1)\,\alpha_2(R'-a)\,1_{R'\ge a-1}$.

2880: By~\eref{e.qm}, this is at most

2881: $O(1)\,\alpha_2(R)\,\alpha_2(R'-a,R)^{-1}\,1_{R'\ge a-1}$.

2882: The probability that $|R'-a|\le 2^j$ is at most $O(1)\,2^{j}/R$.

2883: We also know that $\alpha_2(r_1,r_2)^{-1}\le O(1)\,(r_2/r_1)^{1-\eps'}$

2884: for some $\eps'>0$, by Reimer's inequality~\cite{Reimer} and~\eref{e.fivearm}.

2885: It follows that the probability that such an edge is examined is 

2886: $O(1)\,\alpha_2(R)$.

2887: The $r^{o(1)}$ factor in Theorem~\ref{t.annulus}

2888: is easily replaced by an $O(1)$ factor, if we use Proposition~\ref{p.qm}

2889: in the course of the proof.

2890: Then we get~\eref{e.zz} for the square grid, but without the $r^{o(1)}$ factor.

2891: We may then choose the dependence

2892: between $r$ and $t$ such that

2893: $\alpha(r)\approx t\,r^{\eps/2}$, where $\eps$ is the constant

2894: in~\eref{e.exponeed}. The rest is immediate from~\eref{e.zz}, since

2895: clearly $r^{-\eps/2}\le O(1)\,t^{\eps'}$ for some $\eps'>0$.

2896: 

2897: A consequence of this argument, which applies without assuming~\eref{e.exponeed},

2898: is that for bond percolation on $\Z^2$ we have

2899: $$

2900: \Pb{V_{t,R}\cap V_{0,R}} \le O(t^{-1})\,\Pb{V_{t,R}}^2.

2901: $$

2902: This gives yet another illustration as to how close the result for

2903: $\Z^2$ seems to be --- if the $t^{-1}$ term was improved to

2904: $t^{-1+\eps}$, that would have been enough.

2905: Consequently, significant improvements in the algorithm or in~\eref{e.noise}

2906: would also be sufficient.

2907: 

2908: 

2909: 

2910: 

2911: \section{Some open questions} \label{sec:questions}

2912: 

2913: Following are a few questions and open problems suggested by the

2914: present paper.

2915: 

2916: \begin{enumerate}

2917: \item For the results in Theorems \ref{th:wedges} and \ref{th:cones},

2918: what is the Hausdorff dimension of the set of exceptional times 

2919: in question? We tend to believe that the answer is the upper bound.

2920: In particular, is the upper bound of $31/36$ in

2921: Theorem \ref{th.hd} the correct answer?

2922: \item Prove that there exist exceptional times for percolation

2923: on the square lattice. (See Section~\ref{sec:square} for a discussion.)

2924: \item What is the best $\gamma$ for which Theorems~\ref{th:crossingquantsquare}

2925:  and~\ref{th:crossingquant} hold?

2926: \item

2927: What is the best revealment of an algorithm determining

2928: the event $\cro_R$ in Theorem~\ref{t.square}?

2929: \item

2930: What is the sharp form of Theorem~\ref{t.noise}?

2931: \item

2932: What are the properties of the infinite cluster at an exceptional

2933: time at which it exists? 

2934: For example, what is the growth rate of the number of vertices

2935: in the Euclidean disk of radius $r$ around the origin which

2936: belong to the cluster of the origin at the first time

2937: $t\ge 0$ in which the cluster is infinite?

2938: Is the growth rate the same at all exceptional times?

2939: \item

2940: What is the relationship between the exceptional infinite

2941: cluster and the incipient infinite cluster?

2942: \end{enumerate}

2943: 

2944: \appendix

2945: \section{Appendix: Quasi-multiplicativity}\label{s.appendix}

2946: 

2947: In this appendix, we discuss the $k$-arm probabilities

2948: and prove that they satisfy the corresponding analogue of 

2949: the relation~\eref{e.multip}.

2950: For $R>0$, let $H_R$ be the union  of the hexagons intersecting $B(0,R)$.

2951: For $R>r>0$ let $A_j(r,R)$ denote the event that

2952: there are at least $j$ crossings from $\p H_r$ to $\p H_R$, of alternating colors.

2953: The following result refers to critical site percolation on the triangular

2954: grid and critical bond percolation on the square grid.

2955: 

2956: \begin{proposition} \label{p.qm}

2957: Let $j>0$ be even. There is a constant $C$, depending only on $j$,

2958: such that for all $r<r'<r''$

2959: \begin{equation}\label{e.qm}

2960: C^{-1}\, \Pb{A_j(r,r'')}\le

2961: \Pb{A_j(r,r')}\,\Pb{A_j(r',r'')}\le C\, \Pb{A_j(r,r'')},

2962: \end{equation}

2963: and $\Pb{A_j(r,2\,r)}>1/C$ if $\Pb{A_j(r,2\,r)}>0$

2964: (i.e., if $r$ is large enough to allow $j$ crossings).

2965: Moreover, a corresponding statement holds for critical bond percolation on the

2966: square grid which alternate between primal and dual crossings.

2967: \end{proposition}

2968: 

2969: This theorem would have been a useful tool in~\cite{SW},

2970: had it been available. Instead, the authors of that paper 

2971: proved a weaker form of this which was good enough for their

2972: purposes. Our proof below uses techniques from~\cite{K2},\cite{LSWup2}

2973: and \cite{SW}.

2974: Indeed, the entire results of the appendix do follow from the ideas

2975: of~\cite{K2}. We include them here for completeness, and for ease

2976: of reference. Additionally, though the basic ideas are the same,

2977: in several respects our treatment is a bit different from~\cite{K2}.

2978: 

2979: Below, we will work in the setting of the triangular grid.

2980: The proof for the square grid is essentially the same.

2981: In the setting of the triangular grid, there is the color exchange trick~\cite{KSZ,ADA},

2982: which shows that the probability for having alternating crossings is comparable to the

2983: probability of any color sequence as long as both colors are present.

2984: In the setting of the square grid, as far as we know, such a trick does not exist.

2985: At the end of the appendix we will explain how the proof of Proposition~\ref{p.qm}

2986: can be generalized to any color sequence.

2987: 

2988: In the following, an interface

2989: from $\p H_r$ to $\p H_R$ is an oriented simple path in the

2990: hexagonal grid that has one color of hexagons adjacent to it on the

2991: right, and the opposite color adjacent to it on the left.

2992: Thus, it is the common boundary of a black crossing and a white crossing.

2993: 

2994: 

2995: 

2996: For $R>r>1$, consider all the interfaces crossing from $\p H_{r}$ to

2997: $\p H_{R}$, and define $s(r,R)$ to be the least distance between any

2998: pair of endpoints of two interfaces on $\p H_R$.

2999: If there are no interfaces, we take $s(r,R)=\infty$.

3000: Note that $s(r,R)$ is monotone non-increasing in $r$.

3001: This quantity will roughly measure the \lq\lq quality\rq\rq\

3002: of the interfaces; when $s(r,R)$ is comparable to $R$, the interfaces

3003: are well separated, and, as we will see, easier to extend.

3004: 

3005: \begin{lemma}\label{l.apart}

3006: For all $a\in(0,1)$, $R>0$, $\delta>0$

3007: $$

3008: \Pb{s(a\,R,R)<\delta\,R}\le C\,\delta^\eps,

3009: $$

3010: where $C=C(a)$ is a constant depending only on

3011: $a$ and $\eps>0$ is an absolute constant.

3012: \end{lemma}

3013: 

3014: The lemma probably follows from \cite[Lemma 2]{K2},

3015: but since the proof is rather short, we include a proof

3016: for completeness.

3017: 

3018: \proof

3019: We prove this in the case $a=1/2$.

3020: The general case is essentially the same.

3021: Let $\alpha\subset\p H_R$ be an arc of diameter 

3022: $R/3$, and let $Y$ be the set of points

3023: in $H_R$ at distance at most $R/3$ from

3024: $\alpha$.

3025: Let $\alpha_1$ be one of the two

3026: arcs in $\p Y\cap\p H_R\setminus\alpha$.

3027: Let $\beta_1,\beta_2,\dots,\beta_k$

3028: be the interfaces crossing

3029: from $\p Y\setminus\p H_R$ to $\alpha$,

3030: ordered so that for $i_1<i_2\le k$,

3031: the interface $\beta_{i_1}$ separates

3032: $\alpha_1$ from $\beta_{i_2}$ in $\closure Y$.

3033: Fix a positive integer $i$

3034: and condition on $i\le k$ and on $\beta_i$.

3035: Let $\hat\beta_i$ denote the union of the hexagons adjacent

3036: with $\beta_i$.

3037: Then the

3038: percolation in the connected component $Y_i$ of

3039: $Y\setminus\hat \beta_i$ separated from $\alpha_1$

3040: by $\beta_i$ remains unbiased.

3041: Suppose that the hexagons in $\hat\beta_i$

3042: adjacent to $Y_i$ are white, say.

3043: Let $z_i$ denote the endpoint of $\beta_i$ on

3044: $\alpha$.

3045: The RSW theorem implies that

3046: there is some constant $\eps>0$ such that with conditioned

3047: probability $1-O(1)\,\delta^\eps$

3048: there is a white crossing 

3049: in $Y_i\setminus B(z_i,\delta\,R)$ from

3050: $\p\hat\beta_i$ to $\p H_R$.

3051: On that event, it is clear that

3052: if $k\ge i+1$, then $|z_i-z_{i+1}|\ge \delta\,R$.

3053: We conclude that

3054: $$

3055: \PB{k\ge i+1,\,|z_i-z_{i+1}|\le\delta\,R\md

3056: k\ge i,\,\beta_i}\le O(1)\,\delta^\eps.

3057: $$

3058: The RSW theorem also implies that there is conditioned probability

3059: bounded away from zero that $k=i$

3060: given $k\ge i$ and $\beta_i$ (this would be guaranteed by

3061: an appropriate crossing in $Y_i$ from $\p\hat\beta_i$

3062: to $\p H_R\setminus\alpha$). Therefore,

3063: $\Pb{k\ge i}\le c^i$ for some constant $c<1$.

3064: Thus, 

3065: \begin{multline*}

3066: \PB{k\ge i+1,\,|z_i-z_{i+1}|\le\delta\,R}

3067: \\

3068: =

3069: \PB{k\ge i+1,\,|z_i-z_{i+1}|\le\delta\,R\md k\ge i}\,

3070: \Pb{k\ge i}= O(1)\,c^i\,\delta^\eps\,.

3071: \end{multline*}

3072: We sum this over all $i=1,2,\dots$, and over an appropriate

3073: covering of $\p H_R$ by $O(1)$ arcs $\alpha$ of diameter $R/3$.

3074: %The lemma follows using the monotonicity of $s(r,R)$ in $r$.

3075: The lemma follows.

3076: \QED

3077: 

3078: 

3079: Next, we prove a statement

3080: saying, roughly, that if the crossings are \lq\lq reasonably good\rq\rq,

3081: then there is a conditioned probability bounded away

3082: from zero that they extend and

3083: the extensions are \lq\lq very good\rq\rq.

3084: 

3085: \begin{lemma}\label{l.improve}

3086: For every $j>0$ even, there is a constant $\bar\delta=\bar\delta(j)>0$

3087: such that 

3088: for every $\delta>0$ there is some constant $c(\delta)>0$, depending only on $\delta$,

3089: such that when $R>r$,

3090: $$%\begin{equation}\label{e.cdel}

3091: \PB{A_j(r,4\,R)\cap\{s(r,4R)>4\,\bar\delta\,R\}\md  A_j(r,R)\cap\{ s(r,R)>\delta\,R\} } > c(\delta)\,.

3092: $$%\end{equation}

3093: \end{lemma}

3094: 

3095: \proof

3096: Set $S=\closure{H_R\setminus H_r}$.

3097: We assume that $A_j(r,R)$ holds and

3098: that $s(r,R)>\delta\,R$.

3099: Let $\gamma_0,\dots,\gamma_{k-1}$  ($k\ge j$) be the collection of all interfaces crossing

3100: from $\p H_r$ to $\p H_R$, in counterclockwise order,

3101: where we choose the indexes so that $\gamma_0$ has white hexagons

3102: on the right hand side. 

3103: (The interfaces are oriented from $\p H_r$ to $\p H_R$.)

3104: In the following, we set $\gamma_i:=\gamma_{i'}$

3105: when $i\notin \{0,1,\dots,k-1\}$ and $i'=i\mod k$.

3106: Set $\Gamma=\bigcup_{i\in \N}\gamma_i$.

3107: How does conditioning on the interfaces $\gamma_0,\dots,\gamma_{k-1}$

3108: affect the percolation process? 

3109: Note that the fact that there are no more than $k$

3110: interfaces means that for each $i\in\N$ there is a white crossing

3111: in $S\setminus\Gamma$

3112: from the right hand side of $\gamma_{2i}$ to

3113: the left side of $\gamma_{2i-1}$ and 

3114: a black crossing in $S\setminus\Gamma$ from the left side of

3115: $\gamma_{2i}$ to the right side of $\gamma_{2i+1}$.

3116: Otherwise, the configuration is unbiased on hexagons

3117: that are not adjacent to these interfaces.

3118: 

3119: Let $z_i$ be the endpoint of $\gamma_i$ on $\p H_R$, $i\in\N$.

3120: For $i=0,1,\dots,{j-1}$, let $w_i$ be a point $\p H_R$ that

3121: is roughly in the center of the counterclockwise arc from

3122: $z_i$ to $z_{i+1}$. Then $|w_i-z_i|\ge\delta\,R/5$,

3123:  and the same is true for $|w_i-z_{i+1}|$.

3124: It is easy to see that there exist disjoint simple paths

3125: $\beta_0,\dots,\beta_{j-1}$ satisfying the following.

3126: (See Figure~\ref{f.betas}.)

3127: (1) Each $\beta_i$ is a path in $\closure {H_{4R}\setminus H_R}$ 

3128: from $w_i$ to a point $w_i'\in\p H_{4R}$.

3129: (2) The points $w_i'$ are roughly equally spaced on

3130: $\p H_{4R}$.

3131: (3) Each $\beta_i\cap H_{2R}$ is contained in the line

3132: through the origin containing $w_i$,

3133: and each $\beta_i\setminus H_{3R}$ is contained

3134: in the line through the origin containing $w_i'$.

3135: (4) The distance from each of these paths to any other

3136: path is at least $c_1\,\delta\,R$,

3137: where $c_1\in(0,1/5)$ is some universal constant.

3138: (5) Each $\beta_i$ has length at most constant

3139:  times $R$, where the constant may depend on $j$.

3140: For each $i\in\{0,1,\dots,j-1\}$ let

3141: $\alpha_i$ be the arc of a circle whose center

3142: is $z_i$, that has $w_i$ as endpoint,

3143: that has the other endpoint on $\gamma_i\cup\gamma_{i+1}$,

3144: that is otherwise disjoint from $\gamma_i\cup\gamma_{i+1}$

3145: and is contained in $S$.

3146: 

3147: \begin{figure}

3148: \SetLabels

3149: \E(.5*.5)$H_R$\\

3150: \T(.5*.98)$\p H_{4R}$\\

3151: \endSetLabels

3152: %\ShowGrid

3153: \centerline{\epsfysize=2.5in%

3154: \AffixLabels{%

3155: \epsfbox{betas.eps}%

3156: }%

3157: }

3158: \begin{caption} {\label{f.betas}The paths $\beta_i$.}

3159: \end{caption}

3160: \end{figure}

3161: 

3162: Let $\hat\beta_i$ be the connected component containing $w_i'$ of

3163: the complement of $\Gamma$ in the $c_1\,\delta\,R/20$ neighborhood

3164: of $\beta_i\cup\alpha_i$.

3165: If $i\in\{0,1,\dots,j-1\}$ is odd, let $F_i$ denote the event that

3166: there is a white crossing

3167: in $\hat\beta_i$ from $\Gamma$ to $\p H_{4R}$.

3168: Similarly, if $i$ is even, let $F_i$ denote the event that

3169: there is a black crossing in $\hat\beta_i$ from

3170: $\Gamma$ to $\p H_{4R}$.

3171: It is easy to see that if $\bigcap_{i=0}^{j-1}F_i$ holds,

3172: then $A_j(r,4R)$ holds as well.

3173: The RSW theorem implies that $\Pb{F_i}$ is bounded from

3174: below (depending on $\delta$) for a percolation process that is unbiased.

3175: But, as we have seen,

3176: the percolation on $S$ between $\gamma_i$ and $\gamma_{i+1}$

3177: is only conditioned on having a crossing of the appropriate

3178: color. By the Harris-FKG inequality, this is positively

3179: correlated with $F_i$. By independence on

3180: disjoint sets, the different $F_i$ are independent

3181: given $\Gamma$

3182: (assuming, as we may, that the distance between the different

3183: sets $\hat\beta_i$ is significantly larger than the scale of the lattice). 

3184: We conclude that for some $c(\delta)>0$,

3185: $$

3186: \PB{A_j(r,4\,R)\md  A_j(r,R)\cap\{ s(r,R)>\delta\,R\} } > c(\delta)\,.

3187: $$

3188: 

3189: 

3190: \begin{figure}

3191: \SetLabels

3192: %

3193: \L(.83*.85)$\p H_{4R}$\\

3194: \E(.01*.01)$\hat\beta_i$\\

3195: \L\B(.39*.7)$\p H_{3.5\,R}$\\

3196: \endSetLabels

3197: %\ShowGrid

3198: \centerline{\epsfysize=2.5in%

3199: \AffixLabels{%

3200: \epsfbox{bent.eps}%

3201: }%

3202: }

3203: \begin{caption} {\label{f.bent}A bent strip connecting with a channel.

3204: Indicated are the leftmost crossing of

3205: the intersection of the bent strip and the channel that connects

3206: to $H_r$ and a reasonably likely crossing from it to $\p H_{4R}$ in the bent strip.}

3207: \end{caption}

3208: \end{figure}

3209: 

3210: Taking care of the condition $s(r,4R)\ge 4\,\bar\delta\,R$ is not too hard.

3211: Suppose that in the above we truncate the paths

3212: $\beta_i$ and the neighborhoods $\hat\beta_i$ by intersecting

3213: them with $H_{3.5\,R}$.

3214: We then condition on the \lq\lq leftmost\rq\rq\

3215: crossing in $\hat\beta_i$.

3216: See Figure~\ref{f.bent}. 

3217: All this takes place within $H_{3.5\,R}$.

3218: The conditional probability that these crossings in

3219: the $\hat\beta_i$'s connect to

3220: $\p H_{4\,R}$ is bounded away from

3221: zero by a function of $j$ only (specifically, not $\delta$).

3222: Thus, Lemma~\ref{l.apart} and the monotonicity

3223: of $s(r,R)$ in $r$ shows that if

3224: $\bar\delta=\bar\delta(j)>0$ is chosen small, with conditional

3225: probability at least $1/2$ we are also likely to have $s(r,4R)\ge4\,\bar\delta\,R$,

3226: as required.

3227: \QED

3228: 

3229: 

3230: 

3231: Set 

3232: $$

3233: f(r,R):=\Pb{A_j(r,R)}, 

3234: \qquad

3235: g_\delta(r,R):=\Pb{A_j(r,R)\cap \{s(r,R)>\delta\,R\}}.

3236: $$

3237: 

3238: 

3239: \begin{lemma}\label{l.plenty}

3240: There is a constant $\CCb(j)>0$, depending only on $j$,

3241: such that for $R\ge 4\,r$

3242: $$

3243: \CCb (j)\, g_{\bar \delta}(r,R)\ge f(r,R)\,,

3244: $$

3245: where $\bar\delta=\bar\delta(j)$ is the constant introduced in

3246: Lemma~\ref{l.improve}.

3247: \end{lemma}

3248: 

3249: \proof

3250: We assume that $f(r,R)>0$.

3251: Let $\delta>0$ be small.

3252: Set $N=\log_4(R/r)$ and let $m=m_\delta$ be the largest

3253: integer in the range $0\le m\le N-1$

3254: such that

3255: $ g_\delta(r,4^{-i}R)\le f(r,4^{-i}R)/2 $

3256: holds for every integer $i$ in the range

3257: $0\le i< m$. 

3258: Lemma~\ref{l.apart} and independence

3259: on disjoint sets imply

3260: $$f(r,R)-g_\delta(r,R)\le C\,\delta^\eps\, f(r,R/4),$$

3261: and repeated applications of this inequality give

3262: \begin{equation}\label{e.fR}

3263: f(r,R)\le (2C)^m\,\delta^{\eps\, m} f\bl(r,4^{-m} R\br)\,.

3264: \end{equation}

3265: We claim that if $\delta$ is a sufficiently small positive constant, then

3266: \begin{equation}\label{e.gd}

3267: f(r,4^{-m}R) \le \CCc(j)\,g_\delta(r,4^{-m}R)

3268: \end{equation}

3269: for some constant $\CCc(j)$ depending only on $j$.

3270: If $m\le N-2$, this follows with $\CCc(j)=2$ from the definition of $m$.

3271: If $N-2<m\le N-1$, then RSW easily implies

3272: $f(r,4^{-m}R)\ge \CCa (j)$ for some constant $\CCa(j)>0$,

3273: and $f(r,4^{-m}R)-g_\delta(r,4^{-m}R)\le C\,\delta^\eps$ by Lemma~\ref{l.apart},

3274: which gives~\eref{e.gd}.

3275: %

3276: On the other hand, repeated application of Lemma~\ref{l.improve}

3277: gives

3278: \begin{equation}\label{e.fullcircle}

3279: c(\delta)\,c(\bar\delta)^{m-1}\,g_\delta(r,4^{-m}R)

3280: \le g_{\bar\delta}(r,R) \le f(r,R)

3281: \,.

3282: \end{equation}

3283: On combining this with~\eref{e.fR} and~\eref{e.gd}, one obtains

3284: $$

3285: c(\bar\delta)\,\CCc(j)\,\bl(2\,C\,\delta^\eps/ c(\bar\delta)\br)^m \ge c(\delta)\,.

3286: $$

3287: We choose $\delta$ sufficiently small so that $\delta^\eps< c(\bar\delta)/(4\,C)$.

3288: Then the above shows that $m$ is bounded by a function of $\delta$ and $j$.

3289: The proof is now completed by combining~\eref{e.fR},~\eref{e.gd}

3290: and~\eref{e.fullcircle}.

3291: \QED

3292: 

3293: 

3294: \proofof{Proposition \ref{p.qm}}

3295: The proof will be given only for the triangular grid,

3296: since the proof in the setting of bond percolation on the

3297: square grid is essentially the same.

3298: As we remarked above, when $\Pb{A_j(r,2\,r)}>0$,

3299: the RSW theorem easily gives $\Pb{A_j(r,2\,r)}>1/C(j)$,

3300: for some $C(j)>0$.

3301: 

3302: We now assume that $r'>8\,r$ and $r''>8\,r'$.

3303: Suppose that $A_j(r,r'/2)\cap\{s(r,r'/2)>\bar\delta\,r'/2\}$

3304: holds. We also assume that the corresponding event

3305: occurs between $\p H_{2r'}$ and $\p H_{r''}$,

3306: but now we require that the interfaces be well

3307: separated on the inner boundary $\p H_{2r'}$,

3308: instead of on the outer boundary.

3309: As in the proof of Lemma~\ref{l.improve},

3310: it is not too hard to see that conditioned on these

3311: events there is probability bounded away from zero

3312: (by a function of $j$)

3313: that the crossings between $\p H_r$ and $\p H_{r'/2}$

3314: will hook up nicely with the crossings between

3315: $\p H_{2r'}$ and $\p H_{r''}$.

3316: Basically, we only need to arrange that the channels

3317: $\hat\beta_i$ for the inner crossings will cross the corresponding

3318: channels of the outer crossings.  

3319: The proof of the right hand inequality in~\eref{e.qm}

3320: now follows from Lemma~\ref{l.plenty} and the corresponding

3321: statement for crossings with interfaces well-separated in the

3322: inner boundary, which is proved in the same way.

3323: If $r''\le8\,r'$, then the right hand inequality

3324: in~\eref{e.qm} is a consequence of Lemmas~\ref{l.improve}

3325: and~\ref{l.plenty}. A similar proof applies when

3326: $r'\le 8\, r$.

3327: 

3328: It now remains to prove the left hand inequality in~\eref{e.qm}.

3329: If $r''<2\,r'$, then $\Pb{A_j(r',r'')}$ is bounded away from

3330: zero (if we assume $\Pb{A_j(r,r'')}>0$)

3331: and we are done since $\Pb{A_j(r,r'')}\le \Pb{A_j(r,r')}$.

3332: Otherwise, we argue that

3333: $\Pb{A_j(r,r'')}\le \Pb{A_j(r,r')}\,\Pb{A_j(2\,r',r'')}$,

3334: by independence on disjoint subsets,

3335: and $\Pb{A_j(r',2\,r')}\,\Pb{A_j(2\,r',r'')}\le C\,\Pb{A_j(r',r'')}$,

3336: by the right hand inequality in~\eref{e.qm}.

3337: Since $\Pb{A_j(r',2\,r')}$ is bounded away from zero, the left hand

3338: inequality now follows.

3339: \QED

3340: 

3341: We now generalize Proposition~\ref{p.qm} to arbitrary sequences

3342: of crossings.

3343: 

3344: 

3345: \begin{proposition}\label{p.gencol}

3346: Let $j\ge 1$ be an integer, and fix a color sequence $X\in\{\mathrm{black},\mathrm{white}\}^j$.

3347: The probabilities for the existence of $j$ disjoint crossings

3348: whose colors match this sequence in counterclockwise order

3349: also satisfy the inequalities in Proposition~\ref{p.qm}.

3350: The corresponding statement also holds

3351: in the setting of critical bond percolation on the square grid.

3352: \end{proposition}

3353: 

3354: \proof

3355: We start with the easier case where all the colors in the sequence $X$

3356: are the same, say black.

3357: Suppose that $r'>2\,r$ and $r''>2\,r'$.

3358: Consider the event that (a) there are $j$ disjoint black crossings

3359: from $\p H_r$ to $\p H_{r'}$ and (b) there are $j$ disjoint

3360: black crossings from $\p H_{r'}$ to $\p H_{r''}$

3361: and (c) there are $j$ disjoint black circuits separating

3362: $\p H_{r'/2}$ from $\p H_{r'}$

3363: and (d) there are $j$ disjoint black circuits

3364: separating $\p H_{r'}$ from $\p H_{2\,r'}$

3365: and (e) there are $j$ disjoint crossings from

3366: $\p H_{r'/2}$ to $\p H_{2\,r'}$.

3367: Note that if we choose any one path in each of (a)--(e),

3368: we can extract from the union a crossing from $\p H_r$ to 

3369: $\p H_{r''}$. To see that we actually have

3370: $j$ disjoint crossings from $\p H_r$ to $\p H_{r''}$, note that if

3371: we remove any $j-1$ hexagons, then there is still one path

3372: remaining in each of (a)--(e), and consequently, there is

3373: still a crossing from $\p H_r$ to $\p H_{r''}$.

3374: Thus, Menger's theorem

3375: (see~\cite{Diestel})

3376: % (see \cite{Diestel}, page 50)

3377: implies that when (a)--(e) all hold

3378: there are $j$ disjoint crossings from $\p H_r$ to 

3379: $\p H_{r''}$. By the Harris-FKG inequality, the events (a)--(e)

3380: are all positively correlated. By RSW, events

3381: (c)--(e) have probabilities bounded away from zero

3382: (assuming that (a) has positive probability).

3383: The inequality corresponding to the right hand inequality in~\eref{e.qm}

3384: now follows. The corresponding left hand inequality,

3385: as well as the cases where $r'\le 2\,r$

3386: or $r''\le 2\,r'$ are now proved as in the proof of Proposition~\ref{p.qm}.

3387: 

3388: We now assume that both colors appear in $X$,

3389: and indicate the adaptations necessary in the proof of

3390: Proposition~\ref{p.qm} to generalize to the present setting.

3391: The quantity $s(r,R)$ needs to be defined slightly differently.

3392: In the modified definition of $s(r,R)$, still only 

3393: interfaces between crossings of opposite colors are considered. 

3394: Suppose that $\gamma_1$ and $\gamma_2$ are two adjacent interfaces

3395: from $\p H_r$ to $\p H_R$, and that $Q$ is the component

3396: of $H_R\setminus (H_r\cup\gamma_1\cup\gamma_2)$ between them.

3397: Let $\dist(z,Z;Q)$ denote the infimal length of a path from $z$ to $Z$

3398: in $\closure Q$.

3399: For $s>0$ set $W(s):=\{z\in Q:\dist(z,\p H_R;Q)\le s\}$.

3400: The {\bf margin} between $\gamma_1$ and $\gamma_2$ is

3401: defined as the supremum of the set of 

3402: $s>0$ such that any path connecting

3403: $\gamma_1$ and $\gamma_2$

3404: in $W(s)$ has length at least $s$.

3405: Now $s(r,R)$ is redefined as the smallest margin

3406: among any two consecutive interfaces.

3407: 

3408: Lemma~\ref{l.apart} is still valid with this new definition of $s(r,R)$.

3409: In fact, the only change needed in the proof is that instead of

3410: looking for a white crossing 

3411: in $Y_i\setminus B(z_i,\delta\,R)$ from

3412: $\p\hat\beta_i$ to $\p H_R$, one looks for

3413: a crossing in $Y_i\setminus B(z_i',2\,\delta\,R)$,

3414: where $z_i'$ is the last point on the arc

3415: $\p Y\cap\p H_R$, directed away from $\alpha_1$,

3416: that has distance at most $\delta\,R$

3417: from $\beta_i$.

3418: (Here, we assume that $\delta<1/100$, say.)

3419: 

3420: We now explain how this new definition facilitates the

3421: obvious analogue of Lemma~\ref{l.improve}.

3422: Indeed, suppose that $\gamma_1$ and $\gamma_2$

3423: are two adjacent interfaces, there are

3424: at least $j_1\le j$ disjoint

3425: black crossings in the sector

3426: $Q$ of $H_R\setminus H_r$ between $\gamma_1$ and $\gamma_2$,

3427: and the margin between $\gamma_1$ and $\gamma_2$ is at least $s$.

3428: Let 

3429: \begin{equation*}\begin{aligned}

3430: Q_i& :=\{z\in W(s): 2\,i\,s/(2\,j_1)\le \dist(z,\p H_R;Q)\le (2\,i+1)\,s/(2\,j_1)\}

3431: \,,\\

3432: Q_i^* &:=\{z\in W(s): 2\,i\,s/(2\,j_1)\le \dist(z,\gamma_1;Q)\le (2\,i+1)\,s/(2\,j_1)\}

3433: \,,

3434: \end{aligned}\end{equation*}

3435: where $W(s)$ is as above.

3436: We may then consider the event that in each $Q_i$, $i=0,1,\dots,j_1-1$,

3437: there is a black crossing from $\gamma_1$ to $\gamma_2$,

3438: and in each $Q_i^*$, $i=0,1,\dots,j_1-1$

3439: there is a black crossing from $\p H_R$ to 

3440: $\p W(s)\setminus (\gamma_1\cup\gamma_2\cup \p H_R)$,

3441: and moreover, the latter crossings continue through well directed

3442: channels all the way to $\p H_{4R}$, as in the proof of

3443: Lemma~\ref{l.improve}.

3444: An application of Menger's theorem,

3445: as in the monochromatic case above, will then guarantee

3446: that at the end $j_1$ disjoint black crossings between $\gamma_1$ and

3447: $\gamma_2$ will extend all the way to 

3448: $H_{4R}$. A compatible construction is applied to

3449: each of the other pairs of adjacent interfaces.

3450: 

3451: Of course, when we condition on the interfaces

3452: $\gamma_0,\gamma_1,\dots,\gamma_{k-1}$, we do not

3453: know how many crossings we will have between each pair of

3454: adjacent interfaces.

3455:  But $k=O(R/s(r,R))$, and so the number of distinct

3456: patterns in which crossings with color sequence type $X$

3457: can occur is bounded by a function of $\delta$ and $j$.

3458: (Specifically, a pattern for $X$ is a specification of

3459: how many crossings are selected between each pair of adjacent interfaces

3460: to make up the sequence of crossings compatible with $X$.

3461: There may very well be additional crossings that are ignored.)

3462: Thus, the most likely pattern given the interfaces

3463: occurs with probability bounded away from zero given that

3464: there are crossings of color sequence $X$

3465: and the construction may be based on this most likely pattern.

3466: The occurrence of this pattern given the interfaces and

3467: the information about the color of hexagons adjacent to

3468: the right hand sides of the interfaces will be a monotone

3469: function in the collection of white hexagons in the regions

3470: between interfaces that have white hexagons on their boundary,

3471: and monotone in black hexagons in the other regions.

3472: Thus, again, the Harris-FKG inequality is applicable.

3473: (We do not want to condition on the exact number

3474: of crossings between two adjacent interfaces, as this is not

3475: a monotone function of the configuration.)

3476: 

3477: Similarly, when we attempt to glue crossings between two

3478: different annuli, we condition on the interfaces, and then

3479: aim for the most likely pattern in each annulus.

3480: These are essentially the only modifications needed in the proof.

3481: \QED

3482: 

3483: \begin{remark}\label{r.qmc}

3484: The analogous statements for critical percolation in cones and

3485: wedges also holds, with similar proofs. The wedge case is, in

3486: fact, easier.

3487: \end{remark}

3488: 

3489: \begin{remark}\label{r.primus}

3490: It is also clear that the above proof shows that if we prescribe

3491: $j$ specific disjoint arcs on $\p H_R$ and require the crossings

3492: from $\p H_r$ to land on

3493: these arcs, with a prescribed color for every arc,

3494: the probability for this event is at least a positive constant

3495: times the probability to have $j$ crossings with this

3496: sequential color pattern, where the constant only depends

3497: on the smallest angle at $0$ subtended by any of the $j$ arcs

3498: (provided that $R$ is not too small, so that each of the arcs

3499: has at least one hexagon unshared with any other arc, say).

3500: \end{remark}

3501: 

3502: As a further application, we prove the following

3503: result about $5$-arm and $6$-arm exponents in $\Z^2$.

3504: 

3505: \begin{corollary}\label{c.56}

3506: Consider critical bond percolation on $\Z^2$.

3507: For $R>r\ge 1$ let $F(r,R)$ denote the event that 

3508: there are five open crossings between distances $r$ and $R$

3509: from $0$ of types primal, primal, dual, primal, dual,

3510: in circular order.

3511: Then

3512: \begin{equation}\label{e.fivearm}

3513:  C^{-1}(r/R)^{2}\le

3514: \Pb{F(r,R)} \le C\,(r/R)^{2},

3515: \end{equation}

3516: where $C>0$ is a universal constant.

3517: Moreover, the probability that there are $6$ alternating

3518: crossings: primal, dual, primal, dual, primal, dual,

3519: between distances $r$ and $R$ is at most

3520: $C\,(r/R)^{2+\eps}$, for some constant $\eps>0$.

3521: The same statement applies to any sequence obtained

3522: by inserting one additional primal or dual

3523: entry to the list (primal, primal, dual, primal, dual).

3524: \end{corollary}

3525: 

3526: This result is essentially due to~\cite[Lemma 5]{KSZ} (in

3527: the context of site percolation on the triangular grid,

3528: though the proof is equally applicable to $\Z^2$).

3529: They omit some of the details,

3530: because the proof is long and the argument is similar to

3531: the proof of~\cite[Lemma 4]{K2}.

3532: Now, we can easily present an essentially complete

3533: and relatively short argument.

3534: 

3535: \proof

3536: We begin with the basic argument from~\cite{KSZ}.

3537: Divide the boundary of the circle $\p B(0,R)$ into

3538: $5$ equal arcs, $A_1,\dots,A_5$, in counterclockwise order.

3539: For concreteness, let's take each $A_j$ as the arc between

3540: angles $(2\,j-1)\,\pi/5$ and $(2\,j+1)\,\pi/5$.

3541: For a vertex $v\in B(0,R)$,

3542: let $F_v$ be the event that

3543: there are primal (open) crossings from $v$ to $A_1,A_3$ and $A_4$ and

3544: dual crossings from dual vertices adjacent to $v$ to $A_2$ and to $A_5$, 

3545: and the primal crossings are disjoint, except at $v$.

3546: (By planarity, it follows that the dual crossings are disjoint.)

3547: We claim that $F_v$ can happen for at most one vertex in

3548: $B(0,R/2)$. Indeed, suppose that $F_v\cap F_u$ holds,

3549: where $v,u$ are vertices in $B(0,R/2)$.

3550: Let $\alpha_i$, $i=1,3,4$ denote some simple primal crossings from $v$

3551: to the arcs $A_i$, which are disjoint, except at $v$.

3552: Similarly, let $\alpha_i'$, $i=1,3,4$, be the corresponding

3553: paths for $u$.

3554: Since $\alpha_1\cup\alpha_3$ separates

3555: $A_2$ from $A_5$ in $B(0,R)$, it is clear that

3556: $u\in \alpha_1\cup\alpha_3$. Similarly,

3557: $u\in\alpha_1\cup\alpha_4$. Since $\alpha_3\cap\alpha_4=\{v\}\subset\alpha_1$,

3558: it follows that $u\in \alpha_1$. Let $\beta_1$ be the arc of $\alpha_1$

3559: from $u$ to $A_1$. Since  $\beta_1\cup\alpha_3'$ is a path from $A_1$ to $A_3$,

3560: it contains $v$ or separates $v$ from $A_2$ or from $A_5$.

3561: The latter two possibilities are ruled out by the dual crossings

3562: to $A_2$ and $A_5$ starting at dual vertices adjacent to $v$.

3563: Thus, $v\in\beta_1\cup\alpha_3'$, and similarly,

3564: $v\in \beta_1\cup\alpha_4'$. But since $\alpha_3'\cap\alpha_4'=\{u\}\subset\beta_1$,

3565: we conclude that $v\in \beta_1$, which implies $u=v$.

3566: 

3567: We now claim that $F:=\bigcup_{v\in B(0,R/2)} F_v$ has probability bounded away from $0$. 

3568: Consider the event that there is a crossing from $A_1$ to $A_4$, and consider the rightmost

3569: such crossing $\ell$ (in the sense that it separates any other crossing from $A_5$).

3570: If there is an open path from $A_3$ to $\ell$, but there is no open path from $A_3$

3571: to $A_1$ disjoint from $\ell$, then $F_w$ will hold, where

3572: $w$ is the first vertex $v$ along $\ell$ (when $\ell$ goes from $A_1$ to $A_4$)

3573: that connects to $A_3$ in the complement of $\ell$.

3574: Thus, we need to show that there is probability bounded away from zero that this

3575: happens with $w\in B(0,R/2)$.

3576: Let $L_1$ be the line passing through the origin and the midpoint of $A_1$.

3577: Let $L_2$ and $L_3$ be lines parallel with $L_1$ at distance $R/20$ and $R/10$

3578: from $L_1$, on the side of $L_1$ that contains $A_5$.

3579: By RSW, there is probability bounded away from zero for the existence of

3580: a dual-open crossing from $A_1$ to $A_4$ in the strip between $L_2$ and $L_3$.

3581: By conditioning on the leftmost such crossing (the one closest to $L_1$),

3582: it is easy to see that there is probability bounded away from

3583: zero that such a dual crossing exists and is also connected to $A_5$ by a

3584: dual-open path.

3585: If moreover we have a primal crossing from $A_1$ to $A_4$ in the strip

3586: between $L_1$ and $L_2$, which happens with

3587: probability bounded away from zero, then the rightmost primal crossing between 

3588: $A_1$ and $A_4$ will be contained in the strip

3589: between $L_1$ and $L_3$.

3590: Conditioned on the latter event and on the latter crossing $\ell$,

3591: it happens with probability bounded away from zero that

3592: there is a primal crossing from $A_3$ to $\ell$ whose endpoint on $\ell$

3593: (which is its only point on $\ell$) is within distance

3594: $R/5$ of the origin and there is a dual crossing from $A_2$ to a dual

3595: vertex adjacent to $\ell$ that is within distance $R/5$ of the origin.

3596: On that event, $F$ holds. Thus, $\Pb{F}$ is bounded away from zero.

3597: 

3598: It is easy to see that the proof

3599: of Proposition~\ref{p.gencol} implies

3600: that $\Pb{F_v}\le O(1)\,\Pb{F_w}$ for $v,w\in B(0,R/2)$.

3601: Since the events $F_v$ are disjoint, and since their sum is of order $1$,

3602: it follows that each $F_v$, $v\in B(0,R/2)$ has probability of order $R^{-2}$.

3603: In particular $R^2\,\Pb{F_0}$ is bounded away from zero and $\infty$.

3604: Now~\eref{e.fivearm} follows from Remark~\ref{r.primus}.

3605: 

3606: The statements regarding the $6$-arm exponent now follow

3607: from Reimer's inequality~\cite{Reimer}. Alternatively, we may also deduce them

3608: from Remark~\ref{r.primus}, as follows. If we fix arcs 

3609: $A_1,\dots,A_6$ in counterclockwise order on the radius $R$ circle, 

3610: where the crossings are required to land,

3611: and we require a primal crossing to $A_1$ and a dual crossing to $A_2$,

3612: then we may condition on the most counterclockwise primal crossing $\gamma$ connecting to $A_1$,

3613: then sequentially on the most clockwise crossings to $A_2,A_3,\dots,A_5$  of the required type.

3614: The conditional probability for having yet another crossing to $A_6$ is 

3615: still bounded by $O(1)\,(r/R)^\eps$, for

3616: some constant $\eps>0$. Now we may apply Remark~\ref{r.primus}, to complete the proof.

3617: \QED

3618: 

3619: \bigskip

3620: \noindent{\bf Acknowledgments}.

3621: We thank Harry Kesten for useful advice.

3622: 

3623: \begin{thebibliography}{AAA}

3624: \addcontentsline{toc}{section}{\refname}

3625: %%\bibliographystyle{halpha}

3626: 

3627: \bibitem{ADA}

3628: Aizenman, M., Duplantier, B. and Aharony, A.

3629: Path crossing exponents and the external perimeter in

3630: 2D percolation. 

3631: {\em Phys. Rev. Let.}  {\bf 83}, (1999), 1359--1362.

3632: 

3633: \bibitem{BKS}

3634: Benjamini, I., Kalai, G. and Schramm, O.

3635: Noise sensitivity of Boolean functions and applications to percolation.  

3636: {\em Inst. Hautes \'Etudes Sci. Publ. Math.}  {\bf 90}, (1999), 5--43.

3637: 

3638: \bibitem{BeSc}

3639: Benjamini, I. and Schramm, O.

3640: Exceptional planes of percolation.  

3641: {\em Probab. Theory Related Fields}, {\bf 111}, (1998), 551--564.

3642: 

3643: \bibitem{BSW}

3644: Benjamini, I., Schramm, O. and Wilson, D. B.

3645: Balanced Boolean functions that can be evaluated so that

3646: every input bit is unlikely to be read.  

3647: {\em STOC '05: Proceedings of the thirty-seventh annual ACM symposium on 

3648: Theory of computing}, (2005), 244--250.

3649: 

3650: \bibitem{BMW}

3651: Berg, J. van den, Meester, R. and White, D. G. Dynamic Boolean

3652: models. {\em Stochastic Process. Appl.}  {\bf 69}, (1997), 247--257.

3653: 

3654: \bibitem{BrSt} Broman, E. I. and Steif J. E.

3655: Dynamical Stability of Percolation for Some Interacting Particle Systems 

3656: and $\epsilon$-Movability, {\em Ann. Probab.},  {\bf 34}, (2006), 539--576.

3657: 

3658: \bibitem{CN} Camia, F. and Newman, C. M.

3659: Two-dimensional critical percolation: the full scaling limit,

3660: {\em Comm. Math. Phys.},  {\bf 268}, (2006), 1--38.

3661: 

3662: \bibitem{Diestel} Diestel, R.  {\em Graph theory.} Springer-Verlag, 

3663: (1997), New York.

3664: 

3665: \bibitem{Eva89}

3666: Evans, S. N.  Local properties of Levy processes on a totally disconnected group,

3667: {\em J. Theoret. Probab.} {\bf 2} (1989), 209--259. 

3668: 

3669: \bibitem{FG} 

3670: Fitzsimmons P.J. and Getoor R.K. 

3671: On the potential theory of symmetric Markov processes.

3672: {\em Math.\ Ann.\ } {\bf 281}, (1988), 495--512.

3673: 

3674: \bibitem{GS} Getoor, R.K. and Sharp, M.J. 

3675: Naturality, standardness, and weak duality for Markov processes.

3676: {\em Zeits.\  Wahr.\ verw.\ Gebiete} {\bf 67}, (1984), 1--62.

3677: 

3678: \bibitem{Grimmett} Grimmett, G.  {\em Percolation.} Second edition,

3679: Springer-Verlag, (1999), New York.

3680: 

3681: \bibitem{HPS} H\"aggstr\"om, O., Peres, Y. and Steif, J. Dynamical

3682: percolation. {\em Ann.\ Inst.\ Henri Poincar\`e, Probab.\ et Stat.}

3683: {\bf 33}, (1997), 497--528. 

3684: 

3685: \bibitem{HS} Hara, T. and Slade, G. Mean field behavior and the lace 

3686: expansion, in {\em Probability Theory and Phase Transitions}

3687: (ed.~G.~Grimmett), 

3688: Proceedings of the NATO ASI meeting in Cambridge 1993, 

3689: (1994), Kluwer.

3690: 

3691: \bibitem{Harris} Harris, T. E.  A lower bound on the critical 

3692: probability in a certain percolation process. {\em Proc. Cambridge Phil.\ Soc.}

3693: {\bf 56}, (1960), 13--20.

3694: 

3695: \bibitem{Haw84}

3696: Hawkes, J. {\em Stochastic analysis and applications (Swansea, 1983)}, 130--154,

3697: Lecture Notes in Math., 1095, Springer, Berlin, 1984.

3698: 

3699: \bibitem{Kahane}

3700: Kahane, J. P. {\em Some random series of functions.} 

3701: Second edition, Cambridge University Press: 

3702: (1985), Cambridge.

3703: 

3704: \bibitem{Kesten} Kesten, H. {\em Percolation theory

3705: for mathematicians,} Birkhauser, (1982), New York.

3706: 

3707: \bibitem{K2} Kesten, H. Scaling relations for 2D-percolation.

3708: {\em Commun.\ Math.\ Phys.} {\bf 109}, (1987), 109--156.  

3709: 

3710: \bibitem{KSZ}

3711: Kesten, H., Sidoravicius, V. and Zhang, Y. 

3712: Almost all words are seen in critical site percolation on the 

3713: triangular lattice.  

3714: {\em Electron.\ J. Probab.}  {\bf 3}, (1998), no. 10, 75 pp. (electronic). 

3715: 

3716: \bibitem{KZ} Kesten, H. and Zhang, Y. 

3717: Strict inequalites for some critical exponents in 2D-percolation.

3718: {\em J. Statist.\ Phys.} {\bf 46}, (1987), 1031--1055.

3719: 

3720: \bibitem{Lawler} Lawler, G. F. 

3721: {\em Conformally invariant processes in the plane.} 

3722: Mathematical Surveys and Monographs, {\bf 114}. 

3723: American Mathematical Society, (2005), Providence, RI.

3724: 

3725: \bibitem{LSW}

3726: Lawler, G., Schramm, O. and Werner, W. 

3727: One-arm exponent for critical 2D percolation  

3728: {\em Electron.\ J. Probab.}  {\bf 7}, (2002), no. 2, 13 pp. (electronic).

3729: 

3730: \bibitem {LSWup2}

3731: Lawler, G., Schramm, O. and Werner, W. 

3732: Sharp estimates for Brownian non-intersection probabilities.

3733: {\em In and out of equilibrium} (Mambucaba, 2000), 113--131,

3734: Progr. Probab., 51, (2002), Birkh\"auser.

3735: 

3736: \bibitem {PSSW}

3737: Peres, Y., Schramm, O., Sheffield, S. and Wilson,D. B.

3738: Random-turn Hex and other selection games,

3739: {\em Amer. Math. Monthly}, May, 2007, to appear.

3740: 

3741: \bibitem{PS} Peres, Y. and Steif, J. E.

3742: The number of infinite clusters in dynamical percolation,

3743: {\em Probab. Theory Related Fields}, {\bf 111}, (1998), 141--165.

3744: 

3745: \bibitem{Reimer} Reimer, D.

3746: Proof of the van den Berg-Kesten conjecture,

3747: {\em Combin. Probab. Comput.},  {\bf 9}, (2000), 27--32.

3748: 

3749: \bibitem{Schramm}

3750: Schramm, O. Scaling limits of loop-erased random walks and uniform spanning trees.  

3751: {\em Israel J. Math.}  {\bf 118}, (2000), 221--288. 

3752: 

3753: \bibitem{Smirnov}

3754: Smirnov, S. 

3755: Critical percolation in the plane: conformal invariance, Cardy's formula, 

3756: scaling limits. {\em C. R. Acad. Sci. Paris S\'er. I Math.}  

3757: {\bf 333},  (2001), 239--244.

3758: 

3759: \bibitem{Smirnov1}

3760: Smirnov, S.

3761: Critical percolation in the plane. I. conformal invariance and Cardy's formula.

3762: II. continuum scaling limit. (long version), preprint.

3763: 

3764: \bibitem{SW}

3765: Smirnov, S. and Werner, W. 

3766: Critical exponents for two-dimensional percolation.  

3767: {\em Math. Res. Lett.}  {\bf 8},  (2001), 729--744

3768: 

3769: \bibitem {Talagrand} Talagrand, M.

3770: Concentration of measure and isoperimetric inequalities in product spaces, 

3771: {\em  Publ. I.H.E.S.}, {\bf 81} (1995), 73--205.

3772: 

3773: 

3774: 

3775: 

3776: \end{thebibliography}

3777: 

3778: \bigskip

3779: \filbreak

3780: \begingroup

3781: {

3782: \small

3783: \parindent=0pt

3784: 

3785: Microsoft Corporation\\

3786: One Microsoft Way\\

3787: Redmond, WA 98052, USA\\ 

3788: {\tt

3789: {http://research.microsoft.com/\string~schramm/}}

3790: 

3791: \bigskip

3792: 

3793: Department of Mathematics \\

3794: Chalmers University of Technology 

3795: and G\"{o}teborg University \\

3796: Gothenburg, Sweden \\

3797: {\tt

3798: {steif@math.chalmers.se} \\

3799: {http://www.math.chalmers.se/$\sim$steif/}}

3800: }

3801: 

3802: 

3803: \filbreak

3804: 

3805: \endgroup

3806: 

3807: 

3808: \end{document}

3809: 

3810: 

3811: 

3812: