1: \documentclass[12pt,a4paper]{article}

2: 

3: \usepackage[latin2]{inputenc}

4: \usepackage[mathscr]{eucal}

5: \usepackage{t1enc}

6: \usepackage{amssymb}

7: \usepackage{amsfonts}

8: \usepackage{amsmath}

9: 

10: \usepackage{amsthm}

11: \usepackage{latexsym}

12: \usepackage[dvips]{color}

13: 

14: \usepackage{graphicx}

15: 

16: \usepackage{bbm}

17: 

18: \newtheorem{Theorem}{Theorem}[section]

19: \newtheorem{Remark}[Theorem]{Remark}

20: \newtheorem{Lemma}[Theorem]{Lemma}

21: \newtheorem{Claim}[Theorem]{Claim}

22: \newtheorem{Proposition}[Theorem]{Proposition}

23: \newtheorem{Conjecture}[Theorem]{Conjecture}

24: \newtheorem{Corollary}[Theorem]{Corollary}

25: 

26: \begin{document}

27: 

28: \title{\Huge

29: {\bf Sharp phase transition \\ and critical behaviour in \\ 2D divide and colour models}}

30: 

31: \author{

32: {\bf Andr\'as B\'alint},

33: \thanks{E-mail: abalint@few.vu.nl}

34: {\bf \, Federico Camia},

35: \thanks{Research supported in part by a VENI grant of the NWO

36: (Dutch Organization for Scientific Research).}\,

37: \thanks{E-mail: fede@few.vu.nl}

38: {\bf \, Ronald Meester}

39: \thanks{Research supported in part by a VICI grant of the NWO

40: (Dutch Organization for Scientific Research).}\,

41: \thanks{E-mail: rmeester@few.vu.nl}\\

42: {\sl Department of Mathematics, Vrije Universiteit Amsterdam}

43: }

44: 

45: \maketitle

46: 

47: \begin{abstract}

48: Consider subcritical Bernoulli bond percolation with fixed parameter $p< p_c$.

49: We define a dependent site percolation model by the following procedure:

50: for each bond cluster, we colour all vertices in the cluster black with

51: probability $r$ and white with probability $1-r$, independently of each other.

52: On the square lattice, defining the critical probabilities for the site model and its dual,

53: $r_c(p)$ and $r_c^*(p)$ respectively, as usual, we prove that $r_c(p)+r_c^*(p)=1$ for

54: all subcritical $p$.

55: On the triangular lattice, where our method also works, this leads to $r_c(p)=1/2$,

56: for all subcritical $p$.

57: On both lattices, we obtain exponential decay of cluster sizes below $r_c(p)$,

58: divergence of the mean cluster size at $r_c(p)$, and continuity of the percolation

59: function in $r$ on $[0,1]$.

60: We also discuss possible extensions of our results, and formulate some natural conjectures.

61: Our methods rely on duality considerations and on recent extensions of the classical

62: RSW theorem.

63: \end{abstract}

64: 

65: \noindent {\bf Keywords:} dependent percolation, sharp phase transition,

66: critical behaviour, duality, DaC model, RSW theorem, $p_c=1/2$.

67: 

68: \noindent {\bf AMS 2000 Subject Classification:} 60K35, 82B43, 82B20

69: 

70: 

71: \section{Introduction}

72: 

73: \subsection{Definition of the model and main results}

74: 

75: Despite the vast literature on two-dimensional percolation and the tremendous progress made in

76: its analysis since its introduction as a mathematical theory in \cite{BH}, the exact value of

77: the critical density is known only for a handful of models.

78: The latter cases are typically Bernoulli (\textit{independent}) percolation models endowed with

79: certain \textit{duality} properties, which play a crucial role in the determination of the critical

80: point.

81: In this paper, we will be concerned with the study of (the value of) the critical point and the

82: ``phase diagram'' of certain two-dimensional \textit{dependent} percolation models.

83: 

84: Our main object of interest is the two-dimensional \textit{Divide and Color (DaC) model} introduced

85: by H\"aggstr\"om~\cite{DaC}.

86: For our purposes, it will be sufficient to consider the simplest version of the model, which can be

87: described as follows.

88: Given a graph $G=(\mathcal{V},\mathcal{E})$ with vertex set

89: $\mathcal{V}$ and edge set $\mathcal{E}$, assign to each edge $e \in \mathcal{E}$ value $1$

90: (present/open) with probability $p$ and value $0$ (absent/closed) with probability $1-p$, independently

91: of all other edges.

92: Denote the resulting $\{0,1\}^{\mathcal{E}}$-valued configuration by $Y$, and the corresponding distribution

93: by $\nu_p$.

94: Call \textit{$p$-clusters (``protoclusters'')} the maximal connected components of the graph obtained

95: by removing from $G$ all the closed edges. Next, colour the vertices of each $p$-cluster black with

96: probability $r$ and white with probability $1-r$, independently of all other $p$-clusters.

97: Denote by $X$ the resulting configuration of black and white vertices, by $\mathbb{P}_{p,r}$ the

98: corresponding distribution, and call \textit{$r$-clusters} the maximal connected

99: (via the edge set $\mathcal{E}$) white and black subsets of the vertex set $\mathcal{V}$.

100: These are for us the ``real clusters'', whose percolation properties we are interested in.

101: Indeed, we will be mainly interested in the percolation properties of $X$ for fixed density $p$ of

102: open edges, and will consider the edge configuration $Y$ as an auxiliary object, needed to define $X$.

103: We will in fact argue later in the introduction that our results should still be valid if the

104: product measure $\nu_p$ on configurations of edges -- corresponding to Bernoulli bond percolation --

105: is replaced by some other measure with analogous properties of translation invariance and

106: ergodicity -- e.g., the random cluster measures of which Bernoulli bond percolation is a special

107: case. Nonetheless, our arguments make substantial use of properties of the product measure

108: $\nu_p$, and cannot be immediately applied to models defined using other measures.

109: 

110: We first restrict our attention to the square lattice, with vertex set ${\mathbb Z}^2$ and edge set

111: $\mathcal{E}^2$

112: given by the edges between nearest neighbour elements of ${\mathbb Z}^2$, and its matching graph,

113: with vertex set ${\mathbb Z}^2$ and edge set given by the edges between nearest and next-nearest

114: neighbour elements of ${\mathbb Z}^2$ (i.e., the previous graph with two edges added to each

115: face of the graph along the two diagonals -- see Figure \ref{duallat}).

116: \begin{figure}[h]

117: \centering

118: \includegraphics[scale=0.25, trim= 0mm 0mm 0mm 0mm, clip ]{dualmerpo.EPS}

119: \caption{Part of the matching graph of the square lattice.}

120: \label{duallat}

121: \end{figure}

122: We remind the reader that the measure $\nu_p$ on the square lattice has a percolation phase

123: transition at $p=1/2$~\cite{Kesten}.

124: We denote by $\Theta(p,r)$ the probability that in the DaC model with parameters $p$ and $r$

125: the origin $(0,0)$ of the square lattice is contained in an infinite black $r$-cluster, and by

126: $\Theta^*(p,1-r)$ the probability that it is contained in an infinite white $*$-cluster, where

127: a $*$-cluster is a connected set of vertices of the matching graph of the square lattice (i.e.,

128: connections along the diagonals are allowed).

129: For fixed $p$, we let $r_c(p)=\sup{\{r:\Theta(p,r)=0\}}$ and $r^*_c(p)=\sup{\{r:\Theta^*(p,r)=0\}}$.

130: In \cite{DaC}, Theorem 2.6, it is shown that $r_c(p)$ and $r_c^*(p)$ are non-trivial.

131: For fixed $p$ and $r$, we call the model \textit{critical} if (i) $\Theta(p,r)=0$ and

132: (ii) the mean size of the black $r$-cluster of the origin is divergent (we call the size of a

133: cluster $C$ its cardinality $|C|$, i.e.\ the number of vertices in the cluster).

134: In this context, we have the following results.

135: 

136: \begin{Theorem} (Duality) \label{duality}

137: For all $p<1/2$, $r_c(p)+r_c^*(p)=1$.

138: \end{Theorem}

139: 

140: We remark that van den Berg has recently proved~\cite{Robnew} that this relation holds

141: for a large class of percolation models using methods different from those of this paper.

142: However, the DaC model does not seem to fit in the framework treated in~\cite{Robnew}.

143: 

144: \begin{Theorem} (Exponential decay) \label{exponential-decay}

145: For $p<1/2$ and $r<r_c(p)$, the size of the black $r$-cluster $C_0^r$ of the origin has an exponentially

146: decaying tail, i.e. there exists a constant $c(p,r)>0$ such that

147: \begin{displaymath}

148: \mathbb{P} _{p,r}(|C_0^r|\geq n)\leq e^{-c(p,r)n}

149: \end{displaymath}

150: for all $n\in \mathbb{N}$.

151: \end{Theorem}

152: The proof of this result is quite similar to the proof of Theorem 2 in \cite{Voronoi} (see also the

153: proof of Theorem 5 in \cite{shpthr}), and we do not give it here.

154: 

155: 

156: \begin{Theorem} (Criticality) \label{criticality}

157: The DaC model is critical for $0 \leq p < 1/2$ and $r=r_c(p)$, and for $p=1/2$ and $r \in (0,1)$.

158: It is not critical for $1/2 < p \leq 1$, where $\Theta(p,r)>0$ for all $r>0$.

159: \end{Theorem}

160: 

161: Theorem~\ref{duality} amounts to a duality relation between black percolation and white $*$-percolation.

162: Together with the other two theorems, it provides a complete picture of the phase diagram of the DaC

163: model, summarized below.

164: 

165: \begin{Corollary} (Phase diagram) \label{phase-diagram}

166: \begin{itemize}

167: \item For all $p<1/2$, there exists $r_c(p) \in [1/2,1)$ such that:

168: \begin{enumerate}

169: \item If $r<r_c(p)$, there exists an infinite white $*$-cluster a.s.\ and the size of

170: the black $r$-cluster of the origin has an exponentially decaying tail.

171: \item If $r=r_c(p)$, $\Theta(p,r_c(p))=\Theta^*(p,1-r_c(p))=0$ and the mean size of the black

172: $r$-cluster of the origin and of the white $*$-cluster of the origin are infinite.

173: \item If $r>r_c(p)$, there exists an infinite black $r$-cluster a.s.\ and the size of the

174: white $*$-cluster of the origin has an exponentially decaying tail.

175: \end{enumerate}

176: \item For $p=1/2$, $\Theta(1/2,r)=0$ for all $r \in (0,1)$ and the mean size of the

177: black $r$-cluster of the origin is infinite.

178: \item For all $p>1/2$, $\Theta(p,r)>0$ for all $r \in (0,1]$.

179: \end{itemize}

180: \end{Corollary}

181: 

182: It is interesting to notice that the two regions of the phase diagram where $\Theta(p,r)>0$,

183: namely (1) $0 \leq p < 1/2$ and $r>r_c(p)$, and (2) $1/2 < p \leq 1$ and $r>0$, have different properties.

184: In the first one, there is an infinite black $r$-cluster (somewhere) with probability $1$.

185: This follows from the ergodicity of the measure $\mathbb{P}_{p,r}$ with respect to translations

186: when $p<1/2$ (see Section~\ref{prelim}), and the fact that the event that $x \in {\mathbb Z}^2$

187: belongs to an infinite black $r$-cluster is translation invariant, and has strictly positive probability.

188: In the second region, where $p>1/2$, the probability

189: that there is an infinite black $r$-cluster is bounded away from $1$ for all $r<1$.

190: This also shows that when $p>1/2, r\in(0,1)$, the measure $\mathbb{P}_{p,r}$ is

191: not ergodic with respect to translations (because of the presence of a unique infinite $p$-cluster).

192: 

193: 

194: Another result that follows easily from Theorem~\ref{criticality} is the continuity of the

195: percolation function $\Theta(p,r)$ as a function of $r$ for $p<1/2$.

196: 

197: \begin{Corollary} \label{continuity}

198: For all $p<1/2$, $\Theta(p,r)$ is a continuous function of $r$.

199: \end{Corollary}

200: 

201: The methods used to prove our main results above are not restricted to the square lattice.

202: In particular, they can be applied to the DaC model on the triangular lattice to obtain the

203: following theorem, where $p_c(\mathbb{T}):=2\sin(\pi/18)$ is the critical density for Bernoulli bond

204: percolation

205: on the triangular lattice~\cite{wierman} (see also \cite{se1,se2}).

206: 

207: \begin{Theorem} (Critical point) \label{critical-point}

208: In the context of the DaC model on the triangular lattice, for all $p<p_c(\mathbb{T})$, $r_c(p)=1/2$.

209: \end{Theorem}

210: 

211: We remark that Theorem~\ref{critical-point} looks stronger than Theorem~\ref{duality} because

212: of the self-duality of site percolation on the triangular lattice, with $*$-clusters being of

213: the same nature as $r$-clusters, which immediately implies $r_c(p)=r^*_c(p)$.

214: The proofs of Theorems~\ref{exponential-decay} and~\ref{criticality} and Corollaries~\ref{phase-diagram}

215: and~\ref{continuity} can also be easily adapted to the triangular lattice.

216: 

217: The DaC model discussed in this section is one of the simplest dependent percolation models

218: that one can think of. Indeed, it is defined using only product measures. Nonetheless, despite

219: its simplicity, even results that are by now considered standard for Bernoulli percolation, such

220: as those discussed in this paper, appear to be much harder to prove for the DaC model than in the

221: independent case. The proofs of such results, although based on ideas developed for Bernoulli

222: percolation, require various original arguments that could potentially be helpful in analysing

223: other dependent percolation models.

224: 

225: The structure of the DaC model is very similar in spirit to that of the random cluster model.

226: Following this analogy,

227: the DaC model can be generalised and seen as a particular member of a larger family of models, as

228: explained in the next section.

229: 

230: 

231: \subsection{Other models}

232: 

233: As mentioned just before Theorem~\ref{critical-point}, our methods are robust in the sense that

234: they work, with obvious modifications, on different lattices.

235: Another natural extension of our results would be to replace the product measure $\nu_p$ with

236: other measures.

237: In particular, we have in mind the class of random cluster measures (of which $\nu_p$ is a special

238: case -- see, e.g., \cite{grimmett2}).

239: These are dependent percolation models that unify in a single two-parameter family a variety of

240: stochastic processes of significant importance for probability and statistical physics, including

241: Bernoulli percolation, Ising and Potts models.

242: They are characterized by two parameters, $0 \leq p \leq 1$ and $q>0$, with $q=1$ corresponding

243: to the Bernoulli percolation measure $\nu_p$.

244: For $q \geq 1$, they have positive correlation and are believed to have a phase transition in $p$

245: with exponential decay of correlations for $p<p_c(q)$.

246: For a detailed account on the random cluster model, the reader is referred to~\cite{grimmett2}.

247: 

248: \begin{Conjecture} \label{random-cluster-measures}

249: Let $\tilde X = \tilde X(p,q,r)$ be a configuration generated as in the DaC model but with $\nu_p$

250: replaced by a random cluster measure with parameters $q>0$ and $p<p_c(q)$.

251: For this model, we conjecture that $r_c(p)+r^*_c(p)=1$ on the square lattice and $r_c(p)=1/2$ on the

252: triangular lattice.

253: Furthermore, we conjecture that results analogous to Theorems~\ref{exponential-decay} and~\ref{criticality}

254: and Corollaries~\ref{phase-diagram} and~\ref{continuity} also hold.

255: \end{Conjecture}

256: 

257: One major obstacle in proving the conjecture using the methods of this paper is the lack of a proof

258: of exponential decay of correlations for random cluster measures with $q \neq 1,2$.

259: The case $q=1$ is treated in this paper. The case $q=2$ is particularly interesting since for $r=1/2$

260: it corresponds to the Ising model; we address this case in a forthcoming paper.

261: On the triangular lattice, where self-duality holds, it is easy to see that, at least for $q\geq 2$, the

262: model defined by $\tilde X$ is critical, in our definition of the term, at the purposed critical point.

263: 

264: \begin{Proposition} \label{universality}

265: Let $\tilde X(p,q,r)$ be the model defined above on the triangular lattice with $q\geq 2$, $p<p_c(q)$, and

266: $r=1/2$.

267: Then there is no infinite $r$-cluster a.s.\ but the mean size of the black $r$-cluster of the origin is

268: infinite.

269: \end{Proposition}

270: 

271: Proposition~\ref{universality} implies that $r_c(p) \geq 1/2$ and suggests that indeed $r_c(p)=1/2$.

272: It has a ``universality'' flavour since it suggests that self-duality alone (almost) determines the

273: critical value $r_c(p)=1/2$.

274: One should however compare this result to Theorem~\ref{criticality}, which shows that when $p=p_c$

275: (i.e., either $1/2$ on the square lattice or $2\sin(\pi/18)$ on the triangular lattice) the DaC model

276: has a critical segment, $r \in (0,1)$, rather than a single critical point.

277: The difference between the two cases is that when $p<p_c$ the size of $p$-clusters has an exponentially

278: decaying tail, while at $p_c$ the mean $p$-cluster size diverges and $p$-clusters form circuits around the

279: origin at all scales, turning the percolation of $r$-clusters effectively into a one-dimensional problem.

280: 

281: 

282: \subsection{Scaling limits}

283: 

284: It is natural to ask what the continuum scaling limit (when the lattice spacing is sent to zero)

285: of the DaC model is on the two critical ``curves''

286: (1) $p \in [0,p_c)$, $r=r_c(p)$ and (2) $p=p_c$, $r \in (0,1)$.

287: 

288: For the first critical curve, based on universality considerations, we expect the scaling limit

289: to be the same as for critical Bernoulli site percolation, corresponding to $p=0$ and $r=r_c(0)$.

290: In particular, we expect crossing probabilities to converge to Cardy's formula~\cite{cardy} (as

291: proved by Smirnov~\cite{smirnov} for critical Bernoulli site percolation on the triangular lattice)

292: and the set of all interfaces between black $r$-clusters and white $*$-clusters to converge to the

293: Continuum Nonsimple Loop process described in~\cite{cn1,cn2}.

294: This is in line with the general principle that short range correlations, as produced by the

295: $p$-clusters below $p_c$, do not affect the critical behaviour and the scaling limit.

296: The DaC model in this regime can be seen as Bernoulli site percolation on a random graph whose

297: vertices are the $p$-clusters, and as long as $p<p_c$, the random graph will be, in some sense,

298: ``close'' to the underlying regular lattice.

299: In other words, under the action of the renormalisation group, the critical curve (1) should

300: have a unique fixed point, namely, $p=0$, $r=r_c(0)$.

301: 

302: We expect similar considerations to hold when the product measure $\nu_p$ is replaced by a

303: different random cluster measure (or even in greater generality), and make the following natural

304: conjecture, stated for simplicity for the triangular lattice.

305: 

306: \begin{Conjecture}

307: Let $\tilde X(p,q,r)$ be the model of Proposition~\ref{universality}. For all $q>0$ and all

308: $p<p_c(q)$, the (site percolation) scaling limit of $\tilde X(p,q,1/2)$ is the same as the

309: scaling limit of critical Bernoulli (site) percolation.

310: \end{Conjecture}

311: 

312: In the case of the second critical curve, we expect a different situation,

313: with different scaling limits for different values of $r$.

314: We expect, for instance, that the scaling limit of crossing probabilities will depend on $r$

315: and will not in general be given by Cardy's formula.

316: 

317: 

318: \subsection{Strategy of the proof of Theorem~\ref{duality}}

319: 

320: The proof of our main result, Theorem~\ref{duality}, follows a ``modern'' version

321: (using Russo's formula) of the celebrated proof~\cite{Kesten} by Kesten that the critical

322: probability for Bernoulli (independent) bond percolation on the square lattice is $1/2$

323: (see also~\cite{Russocrpr}).

324: However, since we are dealing with a \textit{dependent} percolation model, the proof requires

325: various modifications, needed for instance to avoid gathering ``too much information.''

326: Also, in this modified version of Kesten's strategy, and due to the dependence

327: structure of the DaC model, we cannot apply the ``traditional'' RSW theorem.

328: We will instead use a recent version of it taken from~\cite{BBB}, which is a strengthened

329: form of the RSW type theorem in~\cite{Voronoi}.

330: 

331: We now describe briefly (and somewhat imprecisely) what one would do in the case of Bernoulli

332: percolation, corresponding to $p=0$. Some notational remarks first:

333: we shall omit the subscript $r$ from the notation of the

334: DaC measure $\mathbb{P}_{p,r}$ and also often write $r_c$ and $r_c^*$ for $r_c(p)$ and $r_c^*(p)$

335: respectively if no confusion is possible.

336: 

337: It follows from standard arguments that $1-r_c^*\leq r_c$.

338: Therefore, the main task is to prove that this inequality is not strict.

339: We proceed by contradiction, assuming that the open interval $(1-r_c^*,r_c)$ is non-empty.

340: Let $V^b_{n,3n}$ denote the presence of a vertical black crossing of an $n \times 3n$

341: rectangle and $H^{w*}_{n,3n}$ the presence of a horizontal white $*$-crossing of the same

342: rectangle (precise definitions will be given in Section~\ref{basic-defs} below).

343: Accordingly, let $H^b_{n,3n}$ denote the event that there is a horizontal black crossing

344: of an $n \times 3n$ rectangle, $V^{w*}_{n,3n}$ that there is a vertical white $*$-crossing

345: of the same rectangle.

346: Since there is a.s.\ no percolation of black vertices for $r \in (1-r^*_c,r_c)$,

347: it is easy to prove that $\limsup_{n \to \infty} {\mathbb P}_r (V^b_{n,3n}) < 1$.

348: But then, by the simple but crucial observation that, no matter how one chooses to colour

349: the vertices inside the $n \times 3n$ rectangle, there is always either a vertical black

350: r-crossing or a horizontal white $*$-crossing, $\liminf_{n \to \infty}{\mathbb P}_r(H^{w*}_{n,3n})>0$

351: for all $r \in (1-r_c^*,r_c)$.

352: This implies that there is a uniform positive probability to have a horizontal white $*$-crossing in

353: the lowest half of an $n \times 8n$ rectangle for all $n$ large enough. (We note that Kesten

354: used squares in his proof. In our case, due to the dependence structure, it will be more

355: convenient to use rectangles. To avoid confusion and prepare the reader for what will

356: come, we employ the same rectangles here.)

357: 

358: Consider the lowest such crossing $\pi_1$ and look for a black vertical $r$-crossing in

359: the left half of the same $n \times 8n$ rectangle from the top of the rectangle to $\pi_1$.

360: Since there is no white $*$-percolation for $r \in (1-r^*_c,r_c)$, with the help of the RSW

361: theorem, one can show that

362: a black vertical crossing from the top of the rectangle to

363: $\pi_1$ exists with probability bounded away from zero.

364: Consider the leftmost such crossing $\pi_2$.

365: Due to the properties of lowest crossings, the presence of such a black crossing

366: implies the presence of a white vertex $x$ on $\pi_1$ which is pivotal for the event

367: $H^{w*}_{n,8n}$.

368: 

369: Next, center at $x$ a sequence of nested annuli intersecting the $n \times 8n$ rectangle.

370: Inside the portion of each annulus intersecting the rectangle and lying above $\pi_1$ and

371: to the right of $\pi_2$, look for a black crossing joining $\pi_2$ with $\pi_1$.

372: Once again, the existence of such crossings with uniform positive probability is assured by the

373: RSW theorem.

374: Every such crossing gives another white vertex on $\pi_1$ which is pivotal for $H^{w*}_{n,8n}$.

375: In this way, choosing the annuli appropriately, one can find many pivotal vertices with high

376: probability.

377: Using Russo's formula it is then possible to conclude that  ${\mathbb P}_r (H^{w*}_{n,8n})$ has

378: a very large (negative) derivative for all $r \in (1-r_c^*,r_c)$, obtaining a contradiction.

379: 

380: The argument above relies on properties of lowest and leftmost crossings, and in

381: particular uses the fact that a lowest (respectively, leftmost) crossing can be

382: found without exploring the area above (resp., to the right of) the crossing itself.

383: In the case of Bernoulli percolation this implies that the configuration above the lowest

384: crossing can be coupled to an independent configuration, and the probability to find a black

385: r-crossing in the left half of the rectangle can be bounded below using the RSW theorem.

386: The same type of argument applies to the portions of annuli to the right of $\pi_2$ and

387: above $\pi_1$, where the configurations can again be coupled to independent configurations,

388: and the probabilities of finding the appropriate crossings bounded below once again

389: using the RSW theorem.

390: 

391: In our case, similar arguments can be used, but the dependence in the model makes them

392: significantly more complex.

393: Moreover, as remarked above, the ``traditional'' RSW theorem cannot be used, and we have

394: to resort to a more recent version~\cite{BBB} which is weaker but more general and, as it

395: turns out, still sufficiently strong for our purposes.

396: 

397: To deal with the dependence structure of the model, in some situations we will ``fatten''

398: certain collections of vertices (e.g., vertices forming a crossing) by adding to them their

399: $p$-clusters. This procedure identifies closed ``barriers'' of edges with the property that

400: colour configurations on different sides of a barrier are conditionally independent

401: (conditioned on the barrier).

402: 

403: We will also use algorithmic constructions carefully designed to explore certain domains

404: looking for monochromatic crossings without obtaining too much information. This will allow

405: us to couple in a useful way the DaC measure ${\mathbb P}_{p,r}$ conditioned on some specific

406: $\sigma$-algebras corresponding to the information obtained while looking for crossings with

407: an unconditional version.

408: 

409: 

410: \subsection{Outline of the paper}

411: 

412: In Section \ref{prelim}, we present the definitions and introduce notation.

413: Then, we collect the tools which are needed to prove the main results.

414: These include known results such as the exponential decay property of subcritical

415: Bernoulli bond percolation,

416: the FKG inequality for the measure $\mathbb{P} _{p,r}$,

417: and the modern RSW theorem from \cite{BBB}.

418: Then we give the natural analogue of Russo's formula for the DaC model

419: (Theorem \ref{tRusso}), and finally state that

420: percolation occurs with positive probability if and only if certain

421: rectangles can be crossed with high probability (Lemma \ref{lessoreq}).

422: 

423: Section \ref{mainproof} contains the proof of Theorem \ref{duality}.

424: In Section \ref{consequences}, we prove Theorem \ref{criticality} and

425: Corollaries \ref{phase-diagram} and \ref{continuity}.

426: Finally, in Section \ref{triang}, we prove our results on the triangular lattice.

427: 

428: \section{Preliminaries} \label{prelim}

429: 

430: \subsection{Basic definitions and notation} \label{basic-defs}

431: 

432: We consider the square lattice, with vertices the points of $\mathbb{Z}^2$, and

433: edges between adjacent vertices (that is, between vertices at Euclidean distance 1).

434: With the usual abuse of notation, we denote both the graph and its vertex set by $\mathbb{Z}^2$,

435: and we write $\mathcal{E}^2$ for the edge set of this graph.

436: 

437: The state space of our configurations is defined as $\Omega :=\Omega _{D}\times \Omega _{C}$,

438: where $\Omega _{D}:=\{0,1\}^{\mathcal{E}^2 }$ corresponds to Bernoulli bond percolation,

439: and $\Omega _{C} :=\{0,1\}^{\mathbb{Z}^2}$ corresponds to colouring.

440: We identify 0 with the colour white, and 1 with black.

441: The probability measure $\mathbb{P}_{p,r}$ is the measure (on the usual $\sigma$-algebra on $\Omega$)

442: obtained by the procedure described in the introduction.

443: 

444: We introduce the set $\tilde\Omega \subset \Omega$ as the set of configurations such that vertices in the same

445: $p$-cluster have the same colour, and we equip $\tilde\Omega $ with a partial ordering as follows.

446: For $\omega _1=(\eta _1,\xi _1), \omega _2=(\eta _2,\xi _2) \in \tilde\Omega $ we say that

447: $\omega _1 \geq \omega _2$ if, for all $x\in \mathbb{Z}^2$, we have $\xi _1 (x)\geq \xi _2(x)$.

448: Note that the ordering depends on the colours of the vertices only, not on the bond configurations. All the

449: configurations in this paper are silently assumed to be in $\tilde{\Omega}$.

450: We call an event $A \subset \tilde{\Omega}$ \textit{increasing} if $\omega \in A$ and

451: $\omega '\geq \omega $ implies $\omega '\in A$.

452: $A$ is a \textit{decreasing event} if $A^c$ is increasing.

453: 

454: We call a sequence of vertices $(x_0,x_1,\ldots ,x_n)$ in $\mathbb{Z}^2$ a \textit{(self-avoiding) path}

455: if for all $i=0,\ldots ,n-1$, $x_i$ and $x_{i+1}$ are adjacent, and for any

456: $0\leq i< j\leq n,\hspace{0.1cm} x_i\neq x_j$.

457: The definition of a \textit{$*$-path} is similar, but for $i=0,\ldots ,n-1$,

458: the vertices $x_i$ and $x_{i+1}$ need to be just \textit{$*$-adjacent} instead of adjacent,

459: which means that their Euclidean distance is 1 or $\sqrt{2}$.

460: A ($*$-)\textit{circuit} is defined in the same way as a ($*$-)path except that $x_n=x_0$.

461: A \textit{horizontal crossing} of a rectangle $R=[a,b]\times [c,d]$, with $a,b,c,d \in {\mathbb Z}$,

462: is a path ${x_0,x_1,\ldots ,x_n}$ such that $x_0\in \{a\}\times [c,d]$, $x_n\in \{b\}\times [c,d]$

463: and for all $i$, $x_i\in R$.

464: A \textit{vertical crossing} of the same rectangle is a path ${x_0,x_1,\ldots ,x_n}$ such

465: that $x_0\in [a,b]\times \{ d\}$, $x_n\in [a,b]\times \{c\}$ and for all $i$, $x_i\in R$.

466: \textit{$*$-circuits}, \textit{horizontal $*$-crossings}, and \textit{vertical $*$-crossings} are defined

467: by replacing paths by $*$-paths in the above definitions.

468: 

469: A \textit{black path} is a path $\pi ={x_0,x_1,\ldots ,x_n}$ such that for all $i=0,\ldots ,n,$ $x_i$

470: is black (i.e. $X(x_i)=1$).

471: \textit{Black circuits}, \textit{black horizontal crossings}, \textit{black vertical crossings} are defined

472: analogously.

473: A \textit{black cluster} is a maximal subset $K$ of $\mathbb{Z}^2$ such that between any two vertices of $K$

474: there exists a black path.

475: The definitions of \textit{white path}, \textit{white circuit}, \textit{white horizontal crossing},

476: \textit{white vertical crossing}, \textit{white cluster} are obtained by replacing black with white.

477: Black and white ($*$-)paths, ($*$-)circuits, ($*$-)crossings, and ($*$-)clusters are defined analogously.

478: 

479: Let $S_{n,m}$ denote the rectangle $[0,n]\times [0,m]$, with $n,m \in {\mathbb N}$.

480: Denote by $V^b_{n,m}$ the event that there is a vertical black crossing in the rectangle $S_{n,m}$;

481: let $H_{n,m}^b$ be the corresponding event with a horizontal crossing.

482: Furthermore, let $B_n^b$ denote the event that there is a black circuit surrounding the midpoint in the

483: annulus $A_n :=S_{3n,3n} \setminus \left( S_{n,n}+\left( n,n\right) \right)$.

484: Here and later, for a set $S$ and a vector $v$, we use the notation $S+v:=\{x:x-v\in S\}$.

485: The analogous events with white crossings are denoted by $V_{n,m}^w$, $H_{n,m}^w$, and $B_n^w$,

486: respectively.

487: A $*$ in the notation will indicate that we are referring to $*$-crossings and $*$-circuits -- for example,

488: $V^{w*}_{n,m}$ denotes the event that there is a vertical white $*$-crossing in $S_{n,m}$.

489: 

490: Let $d$ denote the $L_1$ distance.

491: The distance between two sets of vertices $V_1$ and $V_2$ is defined by $d(V_1,V_2):=\min

492: \{d(x,y)\hspace{0.05cm}:\hspace{0.05cm x\in V_1,y\in V_2}\}$.

493: Let $\partial B(v,n)$ denote the circle of radius $n$ with center at vertex $v$ in the metric $d$,

494: i.e., $\partial B(v,n):=\{w:d(v,w)=n\}$.

495: For a vertex $v\in \mathbb{Z}^2 $, let $C_v ^p$ be the open $p$-cluster of $v$, i.e., the set of

496: vertices that can be reached from $v$ through edges that are open in the underlying Bernoulli bond

497: percolation with parameter $p$.

498: Let us define the \textit{dependence range} of a vertex $v$ by

499: $\mathcal{D}(v):=\max \{n\in \mathbb{N}:C_v ^p\cap \partial B(v,n) \neq \emptyset\}$.

500: 

501: We call an edge set $E=\{e_1,e_2,\ldots ,e_k\}$ a \textit{barrier} if

502: removing $e_1,e_2,\ldots ,e_k$ (but not their end-vertices) separates the graph $\mathbb{Z}^2$ into two

503: or more disjoint connected subgraphs, of which exactly one is infinite.

504: We call the infinite component of $({\mathbb Z}^2,\mathcal{E}^2) \setminus E$

505: the \textit{exterior} of $E$, and denote it by $ext(E)$.

506: We call the union of the finite components the \textit{interior} of $E$, and denote it by $int(E)$.

507: (Note that a barrier as defined above corresponds to a dual circuit in bond percolation. However, since

508: we work with a different sort of duality throughout this paper, we adopt a different term to avoid

509: confusion.)

510: $E=\{e_1,e_2,\ldots ,e_k\}$ is a \textit{closed barrier} if $E$ is a barrier and $e_i$ is closed

511: in the Bernoulli bond percolation (i.e. $Y(e_i)=0$).

512: For a vertex set $A\subset \mathbb{Z}^2$, let $\Delta A$ denote the \textit{edge boundary} of $A$, that is,

513: $\Delta A:=\{(x,y)\in \mathcal{E}^2 : x\in A, y\in \mathbb{Z}^2 \setminus A\}$.

514: Note that for $p < p_c$, the edge boundary of any $p$-cluster is a closed barrier.

515: 

516: \subsection{Preliminary results}

517: \label{prelim-results}

518: 

519: In this subsection we collect several results that are mostly known, follow directly from known results,

520: or can be proved using variations of classical arguments. The exception is Lemma \ref{stochdom}, which

521: is new and very important in the forthcoming construction.

522: 

523: 

524: \begin{Theorem}\emph{(\cite{menshikov},\cite{ab})}

525: \label{lexpdecay}

526: If $p<1/2$, there exists $\psi (p)>0$ such that for all n we have

527: \begin{displaymath}

528: \nu _{p}(\mathcal{D}(0)\geq n)< e^{-n\psi (p)}.

529: \end{displaymath}

530: \end{Theorem}

531: 

532: A short and simple proof of the following classical result is given in Section 4.2 of \cite{shpthr}.

533: 

534: \begin{Lemma}\label{selfdl}

535: The restriction to a rectangle $R$ of any colour configuration $\xi \in \Omega_C$ contains

536: either a black vertical crossing or a white horizontal $*$-crossing of $R$, but never both.

537: In particular, for any $n,m$ we have

538: \begin{displaymath}

539: V_{n,m}^b = \left( H_{n,m}^{w*} \right) ^c.

540: \end{displaymath}

541: \end{Lemma}

542: 

543: The proof of the next result, which shows positive correlation for the DaC model, was obtained

544: by H\"aggstr\"om and Schramm and included in~\cite{DaC}.

545: 

546: \begin{Theorem}\emph{(\cite{DaC})}

547: \label{FKG}

548: Let $A,B$ be increasing events. Then, for any $p,r\in [0,1]$,

549: \begin{displaymath}

550: \mathbb{P}_{p,r}(A\cap B)\geq \mathbb{P}_{p,r}(A)\mathbb{P}_{p,r}(B).

551: \end{displaymath}

552: \end{Theorem}

553: 

554: We shall also need a result which makes precise (and generalizes) the observation that an edge

555: between two vertices of the same colour is more likely to be open than an edge between vertices

556: whose colours are unknown.

557: Let us consider the following scenario for Lemma \ref{stochdom} below:

558: let $B_1, B_2,\ldots ,B_k$ be barriers,

559: $e_1,\ldots ,e_l$ edges in $U:=\bigcup _{i=1} ^kint(B_i)$, $u_1,\ldots ,u_m$ vertices in $U$

560: (where $k,l,m\in \{0,1,2,\ldots \}$), and denote $\bigcap _{i=1} ^kext(B_i)$ by $V$.

561: Fix states $s_i\in \{0,1\}$, $i=1,\ldots ,l$ and colours $c_j\in \{0,1\}$, $j=1,\ldots ,m$.

562: Let $v_1,\ldots ,v_n$ be vertices in $V$ (where $n\in \{0,1,2,\ldots \}$), and let

563: $\kappa \in\{0,1\}$ be a colour.

564: Let $I$ denote the event that $B_1, B_2,\ldots ,B_k$ are closed,

565: $Y(e_i)=s_i$, $X(u_j)=c_j$, and $v_1,\ldots ,v_n$ all have colour $\kappa $.

566: 

567: \begin{Lemma} \label{stochdom}

568: The conditional distribution of the edges in $V$, conditioned on the event $I$ described

569: above, stochastically dominates the measure $\nu _p$.

570: \end{Lemma}

571: 

572: \medskip\noindent

573: {\bf Proof.}

574: We shall prove the lemma with an iteration, determining the states of edges in $V$ one

575: after another.

576: Let $f_1,f_2,\ldots, f_g$ be edges in $V$, and $t_1,t_2,\ldots ,t_g\in \{0,1\}$ states,

577: where $g\in \{0,1,2,\ldots \}$.

578: Let us consider the events $A:=\bigcap_{i=1} ^g\{Y(f_i)=t_i\}$, and $J:=I\cap A$.

579: Take an edge $e$ in $V$ whose state is not determined by $A$.

580: Note that there is no further restriction on the location of $e$: it may be incident on 0, 1 or 2

581: vertices from $\{ v_1,\ldots ,v_n \}$. We shall first show that

582: \begin{equation} \label{eopenJ}

583: \mathbb{P}_{p,r}(\textrm{$e$ open}|J)\geq p.

584: \end{equation}

585: 

586: It is easy to see that (\ref{eopenJ}) is equivalent to

587: $$(1-p)\mathbb{P}_{p,r}(\textrm{$e$ open} , J)\geq p\mathbb{P}_{p,r}(\textrm{$e$ closed} , J),$$

588: which is also equivalent to

589: $$(1-p)\mathbb{P}_{p,r}(J|\textrm{$e$ open})\nu _p(\textrm{$e$ open})\geq p\mathbb{P}_{p,r}(J|\textrm{$e$

590: closed})\nu _p(\textrm{$e$ closed}).$$

591: Since $\nu _p(e$ open$)=p$ and $\nu _p(e$ closed$)=1-p$, it remains to show that

592: \begin{equation}\label{givenstateeJ}

593: \mathbb{P}_{p,r}(J|\textrm{$e$ open})\geq \mathbb{P}_{p,r}(J|\textrm{$e$ closed}).

594: \end{equation}

595: This may be seen as follows. Since the first step in constructing a configuration corresponds to

596: Bernoulli bond percolation, the states of edges other than $e$

597: are independent of the state of $e$. For $J$ to occur, the edges in $\bigcup _{i=1}^kB_i$ need to be closed.

598: In that case, the colouring of $u_1,\ldots ,u_m$ is not influenced by the state of $e$, since every vertex

599: in question is in the interior

600: of one of the closed barriers.

601: Therefore, the only thing left to prove is that the probability that the vertices

602: $v_1,\ldots ,v_m$ all have colour $\kappa $ is greater given that $e$ is open

603: than given that $e$ is closed.

604: This follows immediately from a very simple coupling between $\mathbb{P}_{p,r}(J|\textrm{$e$ open})$

605: and $\mathbb{P}_{p,r}(J|\textrm{$e$ closed})$ in which all the edges except $e$ are in the same

606: state, since when $e$ is open the number of $p$-clusters that need to be assigned colour

607: $\kappa $ is smaller than or equal to the number of $p$-clusters that need to be assigned

608: colour $\kappa $ when $e$ is closed.

609: This observation proves (\ref{givenstateeJ}), finishing the proof of (\ref{eopenJ}).

610: 

611: The full stochastic domination can be shown using (\ref{eopenJ}) iteratively as follows.

612: We shall condition on $I$.

613: Fix a deterministic ordering of the edges in $V$ and use the following iteration:

614: \begin{enumerate}

615: \item Start with $A=\{ 0,1 \}^{\mathcal{E}(V)}$ where $\mathcal {E}(V)$ is the set of edges contained in $V$.

616: \item Determine the state $s_e$ of the first edge $e \in \mathcal {E}(V)$ in the ordering whose state is not

617: yet determined by $A$, according to the conditional distribution $\mathbb{P}_{p,r}( \, \cdot \, | \, I\cap A)$.

618: \item $A:=A\cap \{Y(e)=s_e\}$.

619: \item Go back to step 2.

620: \end{enumerate}

621: It is clear that every edge in $V$ gets a state drawn from the correct

622: distribution after finitely many steps.

623: On the other hand, we know from (\ref{eopenJ}) that for all $A$, the marginal

624: of $\mathbb{P}_{p,r}( \, \cdot \, | \, I\cap A)$ on $\Omega_D$ dominates

625: $\nu_p( \, \cdot \, )$. This proves the desired stochastic domination.

626: \hfill $\Box $\\

627: 

628: \medskip

629: In the proof of Theorem~\ref{duality}, we will use an RSW type theorem that was recently obtained

630: by van den Berg, Brouwer and V\'agv\"olgyi~\cite{BBB}.

631: This is a stronger version of the RSW type theorem used by Bollob\'as and Riordan in~\cite{Voronoi}.

632: Such results are weaker than the classical RSW theorem but more general, and can be applied to models

633: for which the classical RSW theorem has not been proved.

634: We remark that in our proof of $r_c(p)+r_c^*(p)\leq 1$ in Section \ref{mainproof} we seem to need

635: the full strength of the result of van den Berg, Brouwer and V\'agv\"olgyi, as the weaker form

636: proved by Bollob\'as and Riordan does not seem to suffice for our purposes.

637: Stated for the DaC model, the result reads as follows.

638: 

639: \begin{Lemma}

640: \label{RSW}

641: For any $p<1/2,r\in [0,1]$, we have\\

642: \begin{displaymath}

643: \textrm{(a) If }\limsup \limits _{n\to \infty }\mathbb{P}_{p,r}(V_{n,\rho n}^b)>0\textrm{ for some $\rho

644: >0$},

645: \end{displaymath}

646: \begin{displaymath}

647: \textrm{then }\limsup \limits _{n\to \infty }\mathbb{P}_{p,r}(V_{n,\rho n}^b)>0\textrm{ for all $\rho >0$};

648: \end{displaymath}

649: \begin{displaymath}

650: \textrm{(b) If }\limsup \limits _{n\to \infty }\mathbb{P}_{p,r}(V_{n,\rho n}^{w*})>0\textrm{ for some $\rho

651: >0$},

652: \end{displaymath}

653: \begin{displaymath}

654: \textrm{then }\limsup \limits _{n\to \infty }\mathbb{P}_{p,r}(V_{n,\rho n}^{w*})>0\textrm{ for all $\rho

655: >0$}.

656: \end{displaymath}

657: \end{Lemma}

658: 

659: \medskip\noindent

660: {\bf Proof.} First we prove (a).

661: Following~\cite{BBB}, Section 4.3, and~\cite{shpthr}, Section 5.1, it suffices to check conditions

662: (1)--(5) below.

663: For a set $R$ and $\lambda \neq 0$, we write $\lambda R$ for $\{x:x/\lambda \in R\}$. We consider the following

664: five conditions.

665: \begin{enumerate}

666: \item[(1)] For any rectangle $R$, if $h$ and $v$ are a horizontal and a vertical crossing of $R$,

667: respectively, then $d(h,v)\leq 1$.

668: \item[(2)] Increasing events are positively correlated.

669: \item[(3)] The model has the symmetries of $\mathbb{Z}^2$, i.e., is invariant under translations by the

670: vectors $(1,0)$ and $(0,1)$, reflections through the coordinate axes of ${\mathbb Z}^2$, and rotations of

671: 90 degrees.

672: \item[(4)] Disjoint regions are asymptotically independent as we ``zoom out'' (the precise formulation

673: of this condition will be given in Lemma \ref{asind}).

674: \item[(5)] For any fixed rectangle $R$ there is a constant $C$ such that the length of a horizontal

675: crossing of $\lambda R$ is bounded from above by $\lambda^C$ if $\lambda $ is large enough.

676: \end{enumerate}

677: 

678: Condition (1) clearly holds here, since horizontal and vertical black crossings of the same rectangle have

679: at

680: least one vertex in common.

681: Condition (2) is given by Theorem \ref{FKG}.

682: Condition (3) can be checked easily.

683: For condition (4), see Lemma \ref{asind} below.

684: Condition (5) obviously holds, since the model is discrete.

685: 

686: For the proof of (b), the same conditions need to be checked with $*$-crossings instead of crossings

687: in (1) and (5), and increasing events replaced by decreasing events in (2).

688: The new first condition still holds, since even though a horizontal $*$-crossing and a vertical one

689: of the same rectangle do not necessarily have a vertex in common, they are at distance at most 1 from

690: each other.

691: The new second condition, namely that decreasing events are positively correlated, is an easy consequence

692: of Theorem \ref{FKG}, since the complement of a decreasing event is an increasing event.

693: \hfill $\Box $\\

694: 

695: \medskip

696: The next lemma immediately implies weak mixing and ergodicity

697: for the DaC model when $p<p_c$.

698: 

699: \begin{Lemma}

700: \label{asind}

701: Let $p<1/2,r\in [0,1]$. Then for disjoint rectangles $R_1$ and $R_2$,

702: for any $\varepsilon >0$ there exists $\lambda _0>0$ such that for all $\lambda >\lambda _0$,

703: for any events $A_1, A_2$ defined in terms of the colouring of vertices in $\lambda R_1$ and

704: $\lambda R_2$ respectively, we have

705: \begin{displaymath}

706: |\mathbb{P}_{p,r}(A\cap B)-\mathbb{P}_{p,r}(A)\mathbb{P}_{p,r}(B)|\leq \varepsilon .

707: \end{displaymath}

708: \end{Lemma}

709: 

710: \medskip\noindent

711: {\bf Proof.} Fix $p<p_c$.

712: Let $R_1$ and $R_2$ be rectangles at distance $k>0$. Take arbitrary events $A_1, A_2$ defined in terms

713: of the colours of vertices in $R_1$ and $R_2$ respectively.

714: Let $K=K_{R_1,R_2}$ be the event that $R_1$ and $R_2$ are separated by a closed barrier in the

715: bond configuration.

716: If $K$ occurs, then the colours of the vertices in $R_1$ and $R_2$ are conditionally independent.

717: Therefore, $A_1$ and $A_2$ are conditionally independent, conditioned on $K$.

718: The law of total probability, together with the previous observation, gives

719: 

720: \begin{eqnarray}

721: \mathbb{P}_{p,r}(A_1\cap A_2) = \mathbb{P}_{p,r} (A_1|K)\mathbb{P}_{p,r}(A_2|K)\nu _p(K)+\mathbb{P}_{p,r}

722: (A_1\cap A_2|K^c)\nu _p(K^c) \label{A1A2}

723: \end{eqnarray}

724: 

725: Since

726: 

727: \begin{equation}\label{A1}

728: \mathbb{P}_{p,r}(A_1)=\mathbb{P}_{p,r} (A_1|K)\nu _p(K)+\mathbb{P}_{p,r} (A_1|K^c)\nu _p(K^c)

729: \end{equation}

730: and

731: \begin{equation}\label{A2}

732: \mathbb{P}_{p,r}(A_2)=\mathbb{P}_{p,r} (A_2|K)\nu _p(K)+\mathbb{P}_{p,r} (A_2|K^c)\nu _p(K^c),

733: \end{equation}

734: by substituting the right hand sides of equations (\ref{A1A2}), (\ref{A1}) and (\ref{A2}), using the

735: triangle inequality, we obtain

736: 

737: \begin{displaymath}

738: |\mathbb{P}_{p,r}(A_1\cap A_2)-\mathbb{P}_{p,r}(A_1)\mathbb{P}_{p,r}(A_2)|

739: \end{displaymath}

740: \begin{displaymath}

741: \leq |\mathbb{P}_{p,r} (A_1|K)\mathbb{P}_{p,r} (A_2|K)||\nu _p(K)-\nu _p(K)^2|+|\nu _p(K^c)||Q|,

742: \end{displaymath}

743: where $|Q|\leq 4$ since $Q$ is the sum of four products of probabilities.

744: 

745: We also need to notice that if none of the vertices in $R_1\cup R_2$ has a dependence range of at least

746: (say) $\frac{k}{3}$ in the initial random

747: bond configuration Y, then $K$ occurs. Therefore,

748: 

749: \begin{eqnarray*}

750:  \nu _p(K^c)&\leq &  \nu _p\left( \bigcup _{v\in R_1\cup R_2}\{\mathcal{D}(v)\geq \frac{k}{3}\}\right)  \\

751: &\leq &  (|R_1|+|R_2|)\nu _p (\mathcal{D}(0)\geq \frac{k}{3})\\

752: &\leq &(|R_1|+|R_2|)e^{-\psi(p)\frac{k}{3}},

753: \end{eqnarray*}

754: according to Theorem \ref{lexpdecay}.

755: 

756: It immediately follows that for the probability of the event $L:=K_{\lambda R_1,\lambda R_2}$, we have

757: $\nu _p(L^c)\leq (|R_1|+|R_2|)\lambda ^2e^{-\psi(p)\frac{\lambda k}{3}}\to 0$ as

758: $\lambda \to \infty$.\\

759: Fix $\varepsilon >0$, and choose $\lambda _0$ so large that for $\lambda >\lambda _0$, we have $\nu

760: _p(L^c)<\frac{\varepsilon }{8} $.

761: Take arbitrary events $A_1$ and $A_2$, defined in terms of the colours of vertices in $\lambda R_1$ and

762: $\lambda R_2$, respectively.

763: Since $\nu _p(L)\geq 1-\frac{\varepsilon }{8}$ implies $|\nu _p(L)-\nu _p(L)^2|\leq \frac{\varepsilon }{4}$,

764: we obtain

765: \begin{eqnarray*}

766: & &|\mathbb{P}_{p,r}(A_1\cap A_2)-\mathbb{P}_{p,r}(A_1)\mathbb{P}_{p,r}(A_2)|\\

767: & &\leq |\mathbb{P}_{p,r} (A_1|L)\mathbb{P}_{p,r} (A_2|L)|\cdot |\nu _p(L)-\nu _p(L)^2|+|\nu _p(L^c)|\cdot

768: 4\\

769: & &\leq 1\cdot \frac{\varepsilon }{4}+\frac{\varepsilon }{8}\cdot 4\\

770: & & < \varepsilon ,

771: \end{eqnarray*}

772: proving the lemma.

773: \hfill $\Box $\\

774: 

775: \begin{Corollary}

776: \label{erg}

777: For $p<1/2$, the measure $\mathbb{P}_{p,r}$ is weakly mixing and therefore ergodic with

778: respect to translations.

779: \end{Corollary}

780: 

781: We also need a version of Russo's formula~\cite{Russocrpr} (see

782: also~\cite{Grimmett}).

783: Let $A$ be an event, and let $\omega =\left( \eta ,\xi \right) $ be a configuration

784: from $\tilde\Omega$. Let $C$ be an open $p$-cluster from $\eta$.

785: We call $C$ \textit{pivotal} for the pair $\left( A,\omega \right) $ if $I_A(\omega )\neq I_A(\omega ')$

786: where $I_A$ is the indicator function of $A$, $\omega' = (\eta,\xi')$, and $\xi'$ agrees with $\xi$

787: everywhere except that the colour of the vertices in $C$ is different.

788: 

789: \begin{Theorem}

790: \label{tRusso}

791: Let $W$ be a set of vertices with $|W|<\infty $, and let $A$ be

792: an increasing event that depends only on the colours of vertices in $W$.

793: Then we have, for any $p\in [0,1]$,

794: \begin{displaymath}

795: \frac{d}{dr}\mathbb{P}_{p,r}(A)=\mathbb{E}_{p,r}(n(A)),

796: \end{displaymath}

797: where $n(A)$ is the number of $p$-clusters which are pivotal for $A$.

798: \end{Theorem}

799: 

800: \medskip\noindent

801: {\bf Proof sketch.} Let us denote the (finite) set of partitions of the vertices in $W$

802: which are compatible with a bond configuration by $\mathcal{P}_W$, and the (random)

803: partitioning determined by the initial bond percolation by $\mathcal{G}_W$. One can follow the proof

804: of Russo's formula in e.g.\ \cite{MM} to obtain for any $g_W \in \mathcal{P}_W$ that

805: \begin{displaymath}

806: \frac{d}{dr}\mathbb{P} _{p,r}(A|\mathcal{G}_W=g_W)=\mathbb{E}_{p,r}\left( n(A) |\mathcal{G}_W=g_W\right).

807: \end{displaymath}

808: Since

809: \begin{displaymath}

810: \frac{d}{dr}\mathbb{P} _{p,r}(A)=\frac{d}{dr}\sum \limits _{g_W\in \mathcal{P}_W} \mathbb{P}

811: _{p,r}(A|\mathcal{G}_W=g_W)\nu _p(\mathcal{G}_W=g_W).

812: \end{displaymath}

813: and

814: the sum is finite,

815: the sum and the derivative can be interchanged, giving

816: \begin{eqnarray*}

817:  \frac{d}{dr}\mathbb{P} _{p,r}(A)&=& \sum \limits _{g_W\in \mathcal{P}_W} \nu _p(\mathcal{G}_W=g_W)

818: \frac{d}{dr}\mathbb{P} _{p,r}(A|\mathcal{G}_W=g_W)\\

819: &=& \sum \limits _{g_W\in \mathcal{P}_W} \nu _p(\mathcal{G}_W=g_W) \mathbb{E} _{p,r}(n(A)|\mathcal{G}_W=g_W)

820: \\

821: & =&\mathbb{E}_{p,r}(n(A))

822: \end{eqnarray*}

823: \hfill $\Box $\\

824: 

825: 

826: \begin{Corollary}

827: \label{Russocons}

828: If $A$ is a decreasing event depending on colours of vertices in a finite set $W\subset \mathbb{Z}^2$,

829: then we have, for all $p\in [0,1]$,

830: \begin{displaymath}

831: \frac{d}{dr}\mathbb{P}_{p,r}(A)=-\mathbb{E}_{p,r}(n(A)).

832: \end{displaymath}

833: \end{Corollary}

834: 

835: The following lemma gives a finite size criterion for percolation

836: (see \cite{Russocrpr}, Lemma 2).

837: 

838: \begin{Lemma}\label{finitesizecrit}

839: There exists a constant $\varepsilon >0$ which satisfies the

840: following property. If there exists $N\in \mathbb{N}$ such that

841: \begin{equation}\label{littlerange}

842: (N+1)(3N+1)\nu _p(\mathcal{D}(0) \geq \frac{N}{3}) \leq \varepsilon

843: \end{equation}

844: and

845: $$

846: \mathbb{P}_{p,r}(V_{N,3N}^b) >  1-\varepsilon ,

847: $$

848: then $\Theta (p,r)>0$.

849: If there exists $N\in \mathbb{N}$ such that (\ref{littlerange}) holds and

850: $$

851: \mathbb{P}_{p,r}(V_{N,3N}^{w*})> 1-\varepsilon,

852: $$

853: then $\Theta ^*(p,1-r)>0$.

854: \end{Lemma}

855: 

856: We do not give the proof of this lemma here as it uses a well-known

857: coupling argument with a 1-dependent bond percolation model on

858: $\mathbb{Z}^2$ (see, e.g., the proof of Theorem~2.6 in \cite{DaC}

859: or the proof of Theorem~1 in \cite{Voronoi}). Using Theorem~\ref{lexpdecay},

860: Lemma~\ref{finitesizecrit}, and standard arguments, we obtain the

861: following lemma, which relates the occurrence of percolation to the

862: probability of crossing large rectangles.

863: 

864: \begin{Lemma}

865: \label{lessoreq}

866: For $p<1/2$, we have

867: \begin{enumerate}

868: \item[(a)] $\limsup \limits _{n\to \infty } \mathbb{P}_{p,r}(V_{n,3n}^b)=1$ if and only if $\Theta

869: (p,r)>0$.\\

870: \item[(b)] $\limsup \limits _{n\to \infty } \mathbb{P}_{p,r}(V_{n,3n}^{w*})=1$ if and only if $\Theta

871: ^*(p,1-r)>0$.

872: \end{enumerate}

873: \end{Lemma}

874: 

875: In order to state the final result in this section, taken from~\cite{GKR}, we need the following notation.

876: Let $\mu $ be a probability measure on colour configurations where the vertices of $\mathbb{Z}^2$ are each

877: declared black or white.

878: Let us denote the event that the origin is in an infinite black cluster by $0\leftrightarrow \infty $, and

879: the number of infinite black clusters by $N$.

880: 

881: \begin{Theorem}\emph{(\cite{GKR})}

882: \label{GanKeR}

883: Assume that

884: \begin{enumerate}

885: \item[(1)] $\mu $ is invariant under horizontal and vertical translations and axis reflections.

886: \item[(2)] $\mu $ is ergodic (separately) under horizontal and vertical translations.

887: \item[(3)] For any increasing events $E$ and $F$,

888: $$ \mu (E\cap F)\geq \mu (E)\mu (F).$$

889: \item[(4)] $0<\mu (0\leftrightarrow \infty )<1$.

890: \end{enumerate}

891: If assumptions (1)-(4) hold, then

892: $$ \mu (N=1)=1 .$$

893: Moreover, any finite set of vertices is surrounded by a black circuit with probability 1 and, equivalently,

894: all white $*$-clusters are finite with probability 1.

895: \end{Theorem}

896: 

897: 

898: \section{Proof of Theorem \ref{duality}}

899: \label{mainproof}

900: 

901: In this section, we shall prove that for any $p<p_c$ we have $r_c(p)+r_c^*(p)= 1$.

902: This can be split in two parts.

903: The first one is an easy consequence of Theorem~\ref{GanKeR}, stated in the previous section.

904: 

905: \begin{Theorem}

906: \label{groreq}

907: For $p<p_c$, we have $r_c(p)+r_c^*(p)\geq 1$.

908: \end{Theorem}

909: 

910: \noindent \textbf{Proof.} We apply Theorem \ref{GanKeR}. Let us fix $p<p_c$, and assume that

911: $r_c(p)+r_c^*(p)< 1$.

912: Then, we may choose some $r\in (r_c(p),1-r_c^*(p))$. Since $r>r_c(p)$, we have $\Theta (p,r)>0$.

913: On the other hand, it is clear that $\Theta(p,r) \leq r$.

914: This gives $0<\Theta (p,r)<1$, i.e.\ (4) for the measure $\mathbb{P}_{p,r}$.

915: Condition (3) is provided by Theorem \ref{FKG}, (2) by Corollary \ref{erg}, and

916: (1) clearly holds for $\mathbb{P}_{p,r}$. Therefore, all white $*$-clusters

917: are finite with probability 1. However, this cannot be the case since $r<1-r_c^*(p)$.

918: \hfill $\Box $\\

919: 

920: To prove the difficult direction, $r_c(p)+r_c^*(p)\leq 1$, we shall use ideas described in~\cite{Russocrpr},

921: some of which are based on Kesten's proof of $p_c=1/2$ for Bernoulli bond percolation on

922: $\mathbb{Z}^2$ (see \cite{Kesten}). However, here the proof is considerably more difficult

923: due to the dependence structure of the DaC model.

924: Some difficulties are of a geometrical nature, others arise from the fact that we have

925: to use an RSW type theorem which is weaker than the RSW theorem available for Bernoulli

926: (independent) percolation, and used by Kesten~\cite{Kesten} in his celebrated proof.

927: 

928: \begin{Theorem}

929: \label{main}

930: For any $p<p_c$, $r_c(p)+r_c^*(p)\leq 1$.

931: \end{Theorem}

932: 

933: \noindent \textbf{Proof.} We shall prove this theorem by contradiction.

934: Assume that for some $p<p_c$, $r_c(p)+r_c^*(p)> 1$, and fix such a $p$.

935: Most of the time in the rest of the proof, this $p$ will not appear in our notation.

936: By the assumption above, we can choose $1-r_c^* < r_1 < r_2 < r_c$. Since for all

937: $r \in (1-r^*_c,r_c)$, $\Theta (p,r)=0$ and $\Theta ^*(p,1-r)=0$, we have by

938: Lemma~\ref{lessoreq} that

939: \begin{displaymath}

940: \limsup \limits _{n\to \infty } \mathbb{P}_{r_2}(V_{n,3n}^b)<1,

941: \end{displaymath}

942: and

943: \begin{displaymath}

944: \limsup \limits _{n\to \infty } \mathbb{P}_{r_1}(V_{n,3n}^{w*})<1.

945: \end{displaymath}

946: Applying Lemma~\ref{selfdl}, we obtain from these inequalities that

947: \begin{equation}\label{white}

948: \liminf \limits _{n\to \infty } \mathbb{P}_{r_2}(H_{n,3n}^{w*})>0,

949: \end{equation}

950: and

951: \begin{equation}\label{black}

952: \liminf \limits _{n\to \infty } \mathbb{P}_{r_1}(H_{n,3n}^{b})>0.

953: \end{equation}

954: 

955: Inequality (\ref{white}) implies that there exists $\gamma >0$ and an integer $N_0$ such that for

956: all $n>N_0$, we have $\mathbb{P}_{r_2}(H_{n,3n}^{w*})\geq \gamma$.

957: Since $H_{n,3n}^{w*}$ is a decreasing event, by monotonicity this inequality holds in the whole interval:

958: for all $r\in [r_1,r_2]$ and all $n>N_0$,

959: \begin{equation}

960: \label{ivwcr}

961: \mathbb{P}_{r}\left( H_{n,3n}^{w*}\right )\geq \gamma .

962: \end{equation}

963: 

964: Since the measure $\mathbb{P}_{r_1}$ is invariant under 90 degree rotations, Theorem \ref{RSW} and

965: inequality

966: (\ref{black}) imply that

967: \begin{equation}\label{black2}

968: \limsup \limits _{n\to \infty } \mathbb{P}_{r_1}(V_{n,3n}^{b})>0,

969: \end{equation}

970: which implies that there exists $\alpha >0$ and a sequence of side lengths $n_k\to \infty $ as

971: $k\to \infty$ such that for every $k$,

972: \begin{equation}\label{Mfolott}

973: \mathbb{P}_{r_1}(V_{n_k,3n_k}^b)\geq \alpha.

974: \end{equation}

975: 

976: For later purposes we remark that the FKG inequality (Theorem~\ref{FKG}) and a standard pasting

977: argument imply that for each $k$ we have, for all $i \in \mathbb{N}, i \geq 1$,

978: \begin{equation}\label{hosszabbak}

979: \mathbb{P}_{r_1}(V_{n_k,(2i+1)n_k}^b)\geq \alpha ^{2i-1}.

980: \end{equation}

981: Indeed, consider rectangles $[0,n_k]\times [2jn_k,(2j+3)n_k]$ for $j=0,1,\ldots ,i-1$ and squares

982: $[0,n_k]\times [2ln_k,(2l+1)n_k]$ for $l=1,2,\ldots i-1$.

983: If there are black vertical crossings in these $i$ rectangles of size $n_k\times 3n_k$ and black horizontal

984: crossings in the $i-1$ squares of size $n_k\times n_k$, then there is a black vertical crossing in the

985: $n_k\times (2i+1)n_k$ rectangle since horizontal and vertical crossings of the same square meet.

986: Using Theorem~\ref{FKG} and the fact that the probability of a horizontal black crossing in a square

987: is bounded below by the probability of a vertical crossing in an $n_k\times 3n_k$ rectangle,

988: we obtain~(\ref{hosszabbak}).

989: 

990: Let us now fix an integer $L$ with the property that if we consider $L$ Bernoulli (i.e. independent)

991: trials, each with success probability $\alpha ^4/2$, then the probability that there are at least

992: $\frac{32}{(r_2-r_1) \gamma \alpha ^{63}}+1$ successes is at least $1/2$.

993: 

994: Next, we choose an element $m_1$ of the sequence $\{ n_k\}$ (for which~(\ref{Mfolott})

995: holds) so large that it satisfies

996: \begin{equation}\label{sokkicsi}

997: (8m_1^2+8m_1)e^{-\psi (p)\frac{m_1}{100}}\leq \min \left( \frac{\alpha ^4}{2L},\frac{\alpha ^{63}}{4(\alpha

998: ^{63}+1)}\right)

999: \end{equation}

1000: and

1001: \begin{equation}\label{m1nagy}

1002: m_1>\frac{600}{\psi (p)},

1003: \end{equation}

1004: where $\psi (p)$ is the constant corresponding to our fixed $p$ in Theorem \ref{lexpdecay}.

1005: Then take other elements $m_2,m_3,\ldots ,m_L$ in the sequence $\{ n_k\} $ satisfying

1006: \begin{equation}\label{midiff}

1007: m_{i+1}>\frac{151}{49}m_i

1008: \end{equation}

1009: for $i=1,2,\hdots ,L-1$.

1010: Finally, using the constant $N_0$ from (\ref{ivwcr}), we set $N=8n_j$ for some $j$ such that

1011: \begin{equation} \label{Nnagy}

1012: N>\max{(N_0,\frac{9}{2}m_L)}.

1013: \end{equation}

1014: As $N>N_0$, for all $r\in \left[ r_1,r_2\right]$, we have

1015: \begin{equation}\label{whcr}

1016: \mathbb{P}_{r}\left( H_{N,3N}^{w*}\right )\geq \gamma .

1017: \end{equation}

1018: 

1019: Since the annulus $A_n = S_{3n,3n} \setminus (S_{n,n}+(n,n))$ can be split into four overlapping rectangles,

1020: each with sides of length $n$ and $3n$, a standard argument, based on pasting crossings

1021: and the FKG inequality (Theorem~\ref{FKG}) implies that, for $i=1,2,\ldots,L$,

1022: \begin{displaymath}

1023: \mathbb{P}_{r_1}(B_{m_i}^b)\geq \left( \mathbb{P}_{r_1}(V_{m_i,3m_i}^b)\right)^4 .

1024: \end{displaymath}

1025: 

1026: Since $m_1,\ldots,m_L$, and $\frac{N}{8}$ are elements of the sequence $\{ n_k\} $,

1027: we get by (\ref{Mfolott}) and (\ref{hosszabbak}) that $\mathbb{P}_{r_1}(V_{m_i,3m_i}^b)\geq \alpha $

1028: and $\mathbb{P}_{r_1}(V_{\frac{N}{8},8N}^b)\geq \mathbb{P}_{r_1}(V_{\frac{N}{8},65\frac{N}{8}}^b)

1029: \geq \alpha ^{63}$.

1030: By monotonicity these inequalities hold in the whole interval $[r_1,r_2]$.

1031: Hence, for all $r\in \left[ r_1,r_2 \right]$ and for $i=1,\ldots,L$, we obtain

1032: 

1033: \begin{equation}\label{anncross}

1034: \mathbb{P}_{r}(B_{m_i}^b)\geq \alpha ^4,

1035: \end{equation}

1036: and

1037: \begin{equation}\label{sqvercr}

1038: \mathbb{P}_{r}(V_{\frac{N}{8},8N}^b)\geq \alpha ^{63}.

1039: \end{equation}

1040: 

1041: We have now made all the preparation needed for the essential part of the proof.

1042: In the second part, we shall show that there are uniformly many pivotal clusters

1043: for the event $H_{N,8N}^{w*}$, in expectation, in the interval $r\in [r_1,r_2]$.

1044: More precisely, we will show that for all $r\in \left[ r_1,r_2\right] $, we have

1045: \begin{equation}\label{pivotal}

1046: \mathbb{E}_{r}\left( n\left( H_{N,8N}^{w*}\right) \right) > \frac{1}{r_2-r_1},

1047: \end{equation}

1048: where $n\left( H_{N,8N}^{w*}\right)$ denotes the number of $p$-clusters that are

1049: pivotal for the event $H_{N,8N}^{w*}$.

1050: 

1051: Before giving the proof, let us explain how this statement leads to a contradiction.

1052: By putting Corollary \ref{Russocons} and (\ref{pivotal}) together, we obtain

1053: \begin{displaymath}

1054: \max \limits _{r\in \left[ r_1,r_2 \right]}\frac{d}{dr}\mathbb{P}_r(H_{N,8N}^{w*})<-\frac{1}{r_2-r_1}.

1055: \end{displaymath}

1056: However, this cannot be the case since it would imply

1057: \begin{eqnarray}

1058: \mathbb{P}_{r_2 }\left( H_{N,8N}^{w*}\right) & \leq & \mathbb{P}_{r_1}\left( H_{N,8N}^{w*}\right) \

1059:  +(r_2-r_1) \max \limits _{r\in \left[ r_1,r_2 \right]}\frac{d}{dr}\mathbb{P}_r(H_{N,8N}^{w*}) \nonumber \\

1060:  & < & \mathbb{P}_{r_1}\left( H_{N,8N}^{w*}\right) - 1, \nonumber

1061: \end{eqnarray}

1062: which is clearly impossible.

1063: 

1064: Note that it was the assumption that the interval $(1-r_c^*,r_c)$ is non-empty that enabled us to choose

1065: a sub-interval $[r_1,r_2]$ of positive length where the derivative of $\mathbb{P}_r(H_{N,8N}^{w*})$ is

1066: uniformly bounded away from $0$ by $-\frac{1}{r_2-r_1}$.

1067: Since this leads to a contradiction, we conclude that $r_c \leq 1-r_c^*$, as stated in Theorem \ref{main}.

1068: It remains to prove (\ref{pivotal}).\\

1069: 

1070: \noindent \textbf{Proof of inequality (\ref{pivotal}).}

1071: It will be convenient to introduce the following notation to denote certain parts of $S:=S_{N,8N}$.

1072: We will denote by $t(S):=[0,N]\times \{ 8N \}$ the \textit{top of $S$}, by $b(S):=[0,N]\times \{ 0 \}$

1073: its \textit{bottom}, by $l(S):=\{ 0 \} \times [0,8N]$ its \textit{left side}, by

1074: $r(S):=\{ N \} \times [0,8N]$ its \textit{right side}, by $UH(S):=[0,N]\times [4N+1,8N]$ its

1075: \textit{upper half}, and by $LH(S):=[0,\frac{N}{2}]\times [0,8N]$ its \textit{left half}.

1076: 

1077: We shall now present a construction of black and white paths in $S$ which guarantees the existence of

1078: many pivotal clusters, and which succeeds with a high enough probability to provide the desired lower

1079: bound for $\mathbb{E}_{r}\left( n\left( H_{N,8N}^w\right) \right)$.

1080: The construction consists of three parts. In the first part, we show that with probability bounded away

1081: from 0, there is a horizontal white $*$-crossing in the lowest part of $S$.\\

1082: 

1083: \noindent {\bf Part 1.}

1084: We start looking for the lowest white horizontal $*$-crossing of $S$.

1085: It is well-known that the lowest such $*$-crossing can be found (when it exists) by checking only

1086: the colours of vertices (in $S$) below the $*$-crossing and on it.

1087: (The meaning of expressions such as ``below, above, to the right of'' can be made precise via the

1088: Jordan Curve Theorem.)

1089: 

1090: Recall that by (\ref{whcr}), the probability of the event $H^{w*}_{N,3N}$ is uniformly bounded

1091: below by $\gamma$.

1092: Suppose that $H^{w*}_{N,3N}$ occurs.

1093: Denote the lowest horizontal white $*$-crossing in $S_{N,3N}$ by $\Pi _h^{w*}$.

1094: We shall later use the fact that $\Pi _h^{w*}$ is also the lowest horizontal white $*$-crossing in $S$.

1095: So far we have checked sites only below or on $\Pi _h^{w*}$, but not above it.

1096: However, since the model is dependent, we do have some information above $\Pi _h^{w*}$;

1097: for example that the $p$-clusters of the vertices in $\Pi _h^{w*}$ are white.

1098: Therefore, let us consider the thickened $*$-crossing

1099: \begin{displaymath}

1100: \Gamma _h ^{w*}:=\bigcup _{x\in \Pi _h^{w*}}C_x ^p.

1101: \end{displaymath}

1102: 

1103: We denote the portion of $S$ above $\Gamma _h^{w*}$ by $A(\Gamma _h^{w*})$.

1104: We also need to define the following sets (see Figure \ref{rek}):

1105: \begin{eqnarray*}

1106: R_i & := &\left( \left[ (i-1)\frac{N}{8},i\frac{N}{8} \right] \times [0,8N]\right) \cap A(\Gamma _h^{w*})

1107: \textrm{ for } i=1,2,3,4,\\

1108: R_5 & := &\left( \left[ 4\frac{N}{8},7\frac{N}{8} \right] \times [0,8N]\right) \cap A(\Gamma _h^{w*}),\\

1109: R_6 & := &\left( \left[ 7\frac{N}{8},N \right] \times [0,8N]\right) \cap A(\Gamma _h^{w*}),\\

1110: C_{L_1} & := &\bigcup _{x\in S\setminus A(\Gamma _h^{w*})}C_x ^p.

1111: \end{eqnarray*}

1112: 

1113: \begin{figure}[h]

1114: \centering

1115: \includegraphics[scale=0.34, trim= 0mm 0mm 0mm 0mm, clip ]{parparts3.EPS}

1116: \caption{Middle part of the rectangle $S$. The top side of $S_{N,3N}$ is

1117: indicated by a dotted segment. The lower broken line represents the lowest

1118: horizontal white $*$-crossing $\Pi^{w*}_h$ of $S_{N,3N}$. The higher broken

1119: line represents the upper boundary of the thickened $*$-crossing $\Gamma^{w*}_h$.

1120: The dotted lines at the left and right side of $S$ form part of the boundary

1121: of $C_{L_1}$.}

1122: \label{rek}

1123: \end{figure}

1124: 

1125: We know that the edges in the edge boundary $\Delta _1:=\Delta C_{L_1}$ are closed in the underlying

1126: Bernoulli percolation, hence $\Delta _1$ forms a closed barrier.

1127: Note that this barrier is obtained without checking the states of edges or colours of vertices in

1128: $ext(\Delta _1)$.

1129: 

1130: We now claim that with high probability, $\{\Gamma _h ^{w*}\cap UH(S)=\emptyset \}$, and

1131: $\{C_{L_1} \cap R_i=\emptyset \}$ for $i=2,3,4,5$.

1132: Indeed, if all the vertices in $S_{N,3N}$ have a dependence range smaller than $N$,

1133: then the first equality holds.

1134: The only way that any of the latter equalities could be false is that there is a vertex

1135: below $\Gamma _h ^{w*}$ whose $p$-cluster extends above $\Gamma _h ^{w*}$ so much that

1136: it intersects one of the rectangles $R_2,R_3,R_4,R_5$.

1137: For this to happen, there has to be a vertex in $S_{N,3N}$ with a dependence range of at

1138: least $\frac{N}{8}$.

1139: Hence, using crude estimations, we give an upper bound for the probability

1140: that at least one equation is false:

1141: \begin{eqnarray*}

1142: \nu _{p}\left( \bigcup_{x\in S_{N,3N}} \{ \mathcal{D}\left( x \right) \geq \frac{N}{8} \} \right) &

1143: \leq & \nu _{p}\left( \bigcup _{x\in S_{N,3N}} \{ \mathcal{D}\left( x \right) \geq \frac{N}{100} \} \right)

1144: \\

1145: & \leq &\sum \limits _{x\in S_{N,3N}}\nu _{p}\left( \mathcal{D}\left( x \right) \geq \frac{N}{100}\right) \\

1146: & \leq & (N+1)(3N+1) \nu _{p}\left( \mathcal{D}\left( 0 \right) \geq \frac{N}{100}\right) \\

1147: & \leq & (8N^2+8N) e^{-\psi (p) \frac{N}{100}} \\

1148: & \leq & (8m_1^2+8m_1) e^{-\psi (p) \frac{m_1}{100}} \\

1149: & \leq & \frac{\alpha ^4}{2L},

1150: \end{eqnarray*}

1151: by the choice of $m_1$ (see (\ref{sokkicsi})). We have also used Theorem \ref{lexpdecay} and the monotonicity of

1152: $f(x)=(8x^2+8x)e^{-\psi (p)\frac{x}{100}}$ for $x>\frac{600}{\psi (p)}$ (this is justified

1153: because of (\ref{m1nagy})).

1154: Therefore, with probability at least $1-\frac{\alpha ^4}{2L}$,  we have no information about the bond

1155: configuration or the colour of the vertices in $R_2,R_3,R_4,R_5$, so their union provides an

1156: unexplored region in $A(\Gamma _h^{w*})$, which contains $UH(S)$.

1157: 

1158: We now condition on the events $\{\Gamma _h ^{w*}\cap UH(S)=\emptyset \}$, $\{C_{L_1} \cap R_i=\emptyset \}$

1159: for $i=2,3,4,5$, and continue with the second part of our construction.\\

1160: 

1161: \noindent {\bf Part 2.}

1162: In this part, our task is to find the leftmost vertical black path in $R_2\cup R_3\cup R_4$ from $t(S)$

1163: to $\Gamma_h ^{w*}$.

1164: Here and later, if $W$ is a set of white vertices, then by ``a black path to $W$'' we mean a black path

1165: to some vertex at distance 1 from $W$.

1166: Let us consider the following event:

1167: $E_2:=\{$there is a black path from $t(S)$ to $\Gamma_h ^{w*}$ that does not leave $R_2\cup R_3\cup R_4\}$.

1168: Let us denote the $\sigma $-algebra generated by the information we have so far by $\mathcal{F}_c$, and let

1169: us denote the conditional measure $\mathbb{P}_{r}(\cdot |\mathcal{F}_c)$ by $\mathbb{P}_{r}^{(c)}$.

1170: We shall first show that for all $r\in [r_1,r_2]$, we a.s.\ have

1171: \begin{equation}\label{conditioning1}

1172: \mathbb{P}_{r}^{(c)}(E_2)\geq \frac{\alpha ^{63}}{2}.

1173: \end{equation}

1174: 

1175: Let $\omega ^{(c)}=(\eta^{(c)},\xi^{(c)})$ and $\omega =(\eta,\xi)$ be elements of $\Omega$

1176: drawn according to $\mathbb{P}_{r}^{(c)}$ and $\mathbb{P}_{r}$, respectively.

1177: We shall show that $\omega$ and $\omega^{(c)}$ can be coupled in

1178: such a way that if there is no large $p$-cluster in $\eta$ in $R_3$,

1179: and there is a vertical black path in $\xi$ from $t(S)$ to $\Gamma_h^{w*}$

1180: that does not leave $R_3$, then there is a black path

1181: in $\xi^{(c)}$ that does not leave $R_2 \cup R_3 \cup R_4$.

1182: 

1183: We first couple $\eta$ and $\eta^{(c)}$ so that they coincide in the exterior of $\Delta_1$.

1184: This is possible because ${\cal F}_c$ contains information only about $\Delta_1$ and $int(\Delta_1)$,

1185: and Bernoulli percolation configurations restricted to disjoint sets are independent.

1186: Note that the $p$-clusters of $\eta$ can extend beyond $\Delta_1$, therefore each $p$-cluster of

1187: $\eta^{(c)}$ contained in $ext(\Delta_1)$ is a subset of a $p$-cluster of $\eta$, but they are

1188: not necessarily the same.

1189: 

1190: In order to couple $\xi$ and $\xi^{(c)}$, we consider the collection $\mathbb G$ of all

1191: (self-avoiding) paths in $\left[ \frac{N}{4},\frac{3N}{8} \right] \times [0,8N]$ from

1192: $t(R_3):=\left[ \frac{N}{4},\frac{3N}{8} \right] \times \{8N\}$ to $\Gamma_h^{w*}$, and

1193: give them some deterministic order.

1194: We also order the vertices in each path starting from $t(R_3)$ and ending at $\Gamma_h^{w*}$.

1195: We denote the $j$-th vertex of the $i$-th path by $v _i ^j$.

1196: 

1197: To each vertex $x \in {\mathbb Z}^2$ we assign a vector $(c_1(x),c_2(x))$, where $c_1$ and

1198: $c_2$ can take three values: black, white or undefined.

1199: Let ${\cal C}_1$ and ${\cal C}_2$ be the collections of all values assigned respectively to

1200: $c_1(x)$ and $c_2(x)$ for all $x \in {\mathbb Z}^2$, indexed by ${\mathbb Z}^2$.

1201: We start with $c_1(x)$ being undefined for all $x \in {\mathbb Z}^2$, and $c_2(x)$ being the

1202: colour of $x$ given ${\cal F}_c$, or undefined if the colour is not known (note that, in

1203: particular, $c_2(x)$ is undefined for all $x \in ext(\Delta_1)$).

1204: We will generate two coupled colour configurations, $\xi$ and $\xi^{(c)}$, according

1205: to the correct marginal distributions with the help of the following algorithm.

1206: Note that the values assigned to $c_1(x)$ and $c_2(x)$ will change, at least for

1207: some $x$, during the algorithmic construction.

1208: 

1209: Let $c$ be an ``auxiliary'' variable that can take the same three values: black, white and

1210: undefined. We also use two index variables: $i$ and $j$.

1211: 

1212: \begin{enumerate}

1213: \item $i:=1, \, j:=1$.

1214: \item $c:=c_1(v _i ^j)$.

1215: \item

1216: \begin{itemize}

1217: \item If $c=$ black, $j:=j+1$.

1218: \item If $c=$ white, $i:=i+1$ and $j:=1$. Stop if $i>|\mathbb G|$.

1219: \item If $c=$ undefined, with probability $r$, let $c:=$ black, and with probability $1-r$, let $c:=$ white.

1220: Then set $c_1(x):=c$ for all $x\in C^p _{v _i ^j}(\eta )$ (i.e., for all $x$ in the same $\eta$ $p$-cluster

1221: of the current vertex), and $c_2(x):=c$ for all $x\in C^p _{v _i ^j}(\eta ^{(c)})$.

1222: \end{itemize}

1223: \item Stop if ${\cal C}_1$ contains a black path from $t(R_3)$ to $\Gamma_h^{w*}$, otherwise go back to 2.

1224: \end{enumerate}

1225: 

1226: After the algorithm stops, we let $\xi(x)=c_1(x)$ for all $x$'s such that $c_1(x)$ is not undefined,

1227: and $\xi^{(c)}(x)=c_2(x)$ for all $x$'s such that $c_2(x)$ is not undefined.

1228: Note that, because of the nature of the algorithm, the vertices that have not been assigned a colour

1229: are naturally split into $p$-clusters (e.g., if $c_1(x)$ is undefined, then $c_1(y)$ is undefined

1230: for all $y$ in the $\eta$ $p$-cluster of $x$).

1231: We then assign colour black with probability $r$ and white with probability $1-r$ to the $p$-clusters

1232: in $\eta$ and in $\eta^{(c)}$ that have not yet been assigned a colour, independently of each other.

1233: 

1234: We now make three important observations.

1235: 

1236: (1) First of all, it can be easily seen that the configurations $\xi$ and $\xi^{(c)}$ generated in

1237: the way described above are distributed according to the correct distributions, ${\mathbb P}_r$

1238: and ${\mathbb P}_r^{(c)}$ respectively.

1239: 

1240: (2) Moreover, before the very last step of the algorithmic procedure, whenever $c_1(x)$ is black for $x$

1241: in $ext(\Delta_1)$, $c_2(x)$ is also black for that same $x$.

1242: This follows from the fact that, because of the coupling between $\eta$ and $\eta^{(c)}$,

1243: a difference between the $p$-clusters of $\eta$ and those of $\eta^{(c)}$ encountered during

1244: the algorithmic construction can only arise when a $p$-cluster of $\eta$ ``crosses'' $\Delta_1$.

1245: In that case, the $\eta$ $p$-cluster possibly reaches more vertices in $ext(\Delta_1)$ than the

1246: $\eta^{(c)}$ $p$-cluster.

1247: If such an $\eta$ $p$-cluster is coloured white, it makes ${\cal C}_1$ ``more white'' than ${\cal C}_2$.

1248: If it is coloured black, the algorithm stops because a black path from $t(R_3)$ to $\Gamma^{w*}_h$

1249: has been generated.

1250: Therefore, before the very last step of the algorithmic procedure, for every vertex $x$ in $ext(\Delta_1)$

1251: such that $c_1(x)$ is black, $c_2(x)$ is also black, and for every vertex such that $c_2(x)$ is undefined,

1252: $c_1(x)$ is either undefined or white.

1253: This implies that if the algorithm stops because ${\cal C}_1$ contains a black path

1254: from $t(R_3)$ to

1255: $\Gamma^{w*}_h$, then also ${\cal C}_2$ contains a black path from $t(R_3)$ to $\Gamma^{w*}_h$.

1256: 

1257: (3) Finally, at the end of the algorithmic construction described above, $c_2(x)$ can

1258: be black only if $x$ is in $R_3$ or belongs to the $\eta^{(c)}$ $p$-cluster of a vertex

1259: in $R_3$.

1260: 

1261: Now note that, because of the coupling between $\eta$ and $\eta^{(c)}$, if $x \in R_3$ has

1262: dependence range not larger than $N/8$ in $\eta$, the same is true for the range of $x$ in

1263: $\eta^{(c)}$.

1264: Therefore, if no vertex in $R_3$ has a range larger than $N/8$ in $\eta$, when the

1265: algorithm stops because it found a black path in ${\cal C}_1$ from $t(R_3)$ to $\Gamma^{w*}_h$, by the

1266: previous comment and observations (2) and (3) above, there is

1267: a black path in ${\cal C}_2$ from $t(R_3)$ to $\Gamma^{w*}_h$ contained inside

1268: $R_2 \cup R_3 \cup R_4$.

1269: 

1270: It follows that, setting

1271: $$T_{R_3}:=\bigcap _{v\in R_3} \{ \mathcal{D}(v)<\frac{N}{8} \},$$

1272: we obtain a.s.

1273: $$\mathbb{P}_{r}^{(c)}(E_2)\geq \mathbb{P}_{r}\left( V_{\frac{N}{8},8N}^b|T_{R_3}\right )\nu _p(T_{R_3}).$$

1274: Elementary calculations show that

1275: $$\mathbb{P}_{r}\left( V_{\frac{N}{8},8N}^b|T_{R_3}\right )\geq \mathbb{P}_{r}\left( V_{\frac{N}{8},8N}^b\

1276: \right )-\nu _p(T_{R_3}^c),$$

1277: where, with a similar computation as in Part 1,

1278: \begin{eqnarray}

1279: \nu_p(T_{R_3}^c)&=&\nu _p\left( \bigcup _{v\in R_3} \{ \mathcal{D}(v)\geq \frac{N}{8} \} \right) \nonumber

1280: \\

1281: & \leq & (\frac{N}{8} +1)(8N+1) \nu _{p}\left( \mathcal{D}\left( 0 \right) \geq \frac{N}{8} \right) \nonumber

1282: \\

1283: & \leq & (8m_1^2+8m_1) e^{-\psi (p) \frac{m_1}{100}} \nonumber \\

1284: &\leq &\zeta \label{nobigcl},

1285: \end{eqnarray}

1286: where $\zeta=\frac{\alpha ^{63}}{4(\alpha ^{63}+1)}$, and in the last step we used (\ref{sokkicsi}).

1287: As $\mathbb{P}_{r}\left( V_{\frac{N}{8},8N}^b\right ) \geq \alpha ^{63}$ (see (\ref{sqvercr})),

1288: and $\zeta <\frac{\alpha ^{63}}{2(\alpha ^{63}+1)}$, this gives

1289: $$\mathbb{P}_{r}^{(c)}(E_2)\geq (\alpha ^{63}-\zeta )(1-\zeta )\geq \alpha ^{63}(1-\zeta )-\zeta

1290: \geq \frac{\alpha ^{63}}{2},$$ proving (\ref{conditioning1}).

1291: 

1292: Having shown that $\mathbb{P}_{r}^{(c)}(E_2)>0$, conditioning on the event $E_2$, we call

1293: $\tilde\Pi^b_v$ the leftmost black (self-avoiding) path from $t(S)$ to $\Gamma_h^{w*}$

1294: contained in $R_2 \cup R_3 \cup R_4$.

1295: We denote by $\Pi^b_v$ the union of $\tilde\Pi^b_v$ and the black $r$-clusters in $S$ to the left

1296: of $\tilde\Pi^b_v$ connected to $\tilde\Pi^b_v$ (see Figure~\ref{leftm7}).

1297: \begin{figure}[h]

1298: \centering

1299: \includegraphics[scale=0.19, trim= 0mm 0mm 0mm 0mm, clip ]{leftm7.EPS}

1300: \caption{The black vertices joined by a line represent a portion of $\tilde\Pi_v^b$.

1301: The black vertices to the left side of $\tilde\Pi_v^b$ and marked by circles belong

1302: to $\Pi_v^b$.}

1303: \label{leftm7}

1304: \end{figure}

1305: We denote the region of $S$ to the right of $\tilde\Pi_v^b$ by $J(\tilde\Pi_v^b)$.

1306: Note that no information on the colours of vertices to the right of $\tilde\Pi_v^b$

1307: is needed to determine $\tilde\Pi_v^b$ itself and $\Pi_v^b$.

1308: 

1309: We now set

1310: $$C_{L_2}:=\bigcup _{x\in (R_2\cup R_3\cup R_4)\setminus (\Pi _v ^b\cup J(\tilde\Pi _v ^b))}C_x^p.$$

1311: If there is no vertex in $R_2\cup R_3\cup R_4$ on the left side of $\tilde\Pi_v^b$

1312: with a dependence range greater than $N$, then $C_{L_2}$ does not intersect $S_{N,7N}\cap J(\tilde\Pi _v ^b)$.

1313: Therefore, the probability of $\{ C_{L_2}\cap S_{N,7N}\cap J(\tilde\Pi_v^b)\neq \emptyset \}$ is at most

1314: $\frac{\alpha ^4}{2L}$, from a computation similar to the one leading to inequality (\ref{nobigcl}).

1315: Conditioning on the event $\{ C_{L_2}\cap S_{N,7N}\cap J(\tilde\Pi _v ^b)= \emptyset \}$, we continue with the

1316: third part of our construction.\\

1317: 

1318: \noindent{\bf Part 3.} In this part, we shall complete our

1319: construction which shows that there are ``too many'' pivotal

1320: $p$-clusters for having a white horizontal $*$-crossing in $S$. It

1321: is easy to see that if a vertex at distance $1$ from some $x \in

1322: \Gamma^{w*}_h$ is black and is connected to $t(S)$ by a black path

1323: contained in $S$, then $C^p_x$ is pivotal for $H_{N,8N}^{w*}$.

1324: Indeed, due to the fact that every vertex in the lowest horizontal

1325: white $*$-crossing has a black neighbour that is connected to

1326: $b(S)$ by a black path, changing the colour of $C^p_x$ would make

1327: the existence of a white horizontal $*$-crossing of $S$ impossible.

1328: 

1329: Let $\pi$ be the

1330: ``upper layer'' of $\Gamma^{w*}_h$, that is,

1331: the set of vertices in $\Gamma^{w*}_h$ with at least

1332: one neighbour in the region $A(\Gamma^{w*}_h)$.

1333: Denote by $x_0$ the rightmost vertex of $\pi$ at distance 1 from $\Pi _v ^b$.

1334: Furthermore, denote the portion of $\pi$ to the right of $x_0$ by $\Pi_u^{w*}$.

1335: We have specified a site $x_0$ of $\Gamma _h^{w*}$ that lies in $LH(S)\cap S_{N,4N}$

1336: (i.e. the lower-left quarter of $S$) with the property that a black path leads from

1337: $t(S)$ to a point at distance 1 from $x_0$. This implies that $C_{x_0}^p$ is pivotal

1338: for $H_{N,8N}^{w*}$.

1339: 

1340: Recall that we defined $m_1,\ldots,m_L$ so that

1341: (\ref{sokkicsi})--(\ref{midiff}) and (\ref{anncross}) hold.

1342: Now let us consider annuli $B_i:= \{ c_i+S_{3m_i,3m_i}+\left(

1343: -\frac{3m_i}{2},-\frac{3m_i}{2}\right) \} \setminus \{

1344: c_i+S_{m_i,m_i}+\left( -\frac{m_i}{2},-\frac{m_i}{2}\right) \}$,

1345: $i=1,\ldots,L$, centered at $c_i:=x_0$ if $m_i$ is even and at

1346: $c_i=x_0+(1/2,1/2)$ if $m_i$ is odd. This means that

1347: for any $i$, $B_i$ is an annulus with center at distance at most

1348: $1/2$ from $x_0$, with inner diameter $m_i$ and outer diameter

1349: $3m_i$. We will look for black paths in these annuli between $\Pi

1350: _v^b$ and $\Pi _u ^{w*}$. Note that even the largest annulus, $B_L$

1351: does not go above $S_{N,7N}$, nor to the right of $R_5$, according to

1352: (\ref{Nnagy}). Let us denote the bounded region determined by the

1353: curves $\left( \{ 7\frac{N}{8} \}\times [0,8N]\right) \cap A(\Gamma

1354: _h^{w*}) $ (i.e. the right side of $R_5$), $[0,N]\times \{ 7N \}$

1355: (the top of $S_{N,7N}$), $\Pi _v^b$, and $\Pi _u ^{w*}$, by $AJ$.

1356: 

1357: Let $\Delta _2$ denote the edge boundary of $C_{L_2}$, defined at the

1358: end of Part 2.

1359: We shall look for black paths in the annuli in the region $AJ$. Let

1360: $B_i[\frac{m_i}{100}]$ denote the $\frac{m_i}{100}$-neighbourhood of

1361: $B_i$. We consider the events $Q_i, i=1,\ldots ,L$, that there is a

1362: black path in $B_i[\frac{m_i}{100}]\cap AJ$ between $\Pi _v^b$ and

1363: $\Pi _u^{w*}$, and $Q:=\{$there are at least $\frac{32}{(r_2-r_1)

1364: \alpha ^{63} \gamma }+1$ indices $i$ such that $Q_i$ holds$\}$. Let

1365: us denote the $\sigma $-algebra generated by the information we have

1366: so far by $\mathcal{F}_{c_2}$, and let us denote the conditional

1367: measure $\mathbb{P}_{r}(\cdot |\mathcal{F}_{c_2})$ by

1368: $\mathbb{P}_{r}^{(c_2)}$. We shall show that for any $r\in

1369: [r_1,r_2]$, a.s.,

1370: \begin{equation}\label{qwerty}

1371: \mathbb{P}_r^{(c_2)}(Q)\geq \frac{1}{4}.

1372: \end{equation}

1373: 

1374: Let $\omega^{(c_2)}=(\eta^{(c_2)},\xi^{(c_2)})$ and $\omega =(\eta,

1375: \xi)$ be configurations in the plane, drawn according to $\mathbb{P}_{r}^{(c_2)}$

1376: and $\mathbb{P}_{r}$, respectively. We shall show that $\omega $ and

1377: $\omega^{(c_2)}$ can be coupled in such a way that if in $\omega$, for

1378: all $i=1,\ldots,L$, there is no vertex in $B_i$ with a dependence range

1379: larger than $\frac{m_i}{100}$, and for some $j$ there is a black

1380: circuit in $B_j$ in $\xi $, then $\xi ^{(c_2)}\in Q_j$.

1381: 

1382: First, we couple the edge configurations $\eta $ and $\eta ^{(c_2)}$

1383: in $ext(\Delta _1)\cap ext(\Delta _2)$.

1384: $\mathcal{F}_{c_2}$ contains information that the edges in $\Delta _1\cup\Delta _2$

1385: are closed, and $\Pi _v ^b$ is coloured black

1386: (plus some information about $int(\Delta _1)\cup int(\Delta _2)$,

1387: but that has no influence on the exterior).

1388: This implies that, according to Lemma \ref{stochdom}, there will be a

1389: bias in the configuration $\eta^{(c_2)}$ towards more open edges.

1390: In fact, according to Lemma \ref{stochdom}, in $ext(\Delta _1)\cap ext(\Delta _2)$,

1391: $\eta$ and $\eta^{(c_2)}$ can be coupled so that any closed edge in the latter is

1392: also closed in the former. We pick two such coupled configurations, and concentrating

1393: on $\eta^{(c_2)}$ first, we denote by $H$ the union of $\Pi _v^b$ with the set of vertices

1394: that are connected to $\Pi _v^b$ by an open path in $\eta ^{(c_2)}$. Then $\Delta H$ is a

1395: closed barrier in $\eta ^{(c_2)}$, and by the coupling, this barrier is also closed in $\eta$.

1396: As a final ingredient in our joint construction, we now redraw the configurations in

1397: $ext(\Delta _1)\cap ext(\Delta _2)\cap ext(\Delta H)$ in both configurations, so that in

1398: this region the configurations agree, and are (conditionally) independent of

1399: $\Delta _1, \Delta _2,\Delta H$ and their interiors. The configurations chosen this

1400: way, denoted again by $\eta$ and $\eta^{(c_2)}$, have the correct marginal distributions.

1401: 

1402: As in Part 2, we assign a vector $(c_1(x),c_2(x))$ to each vertex $x\in \mathbb{Z}^2$,

1403: and let ${\cal C}_1,{\cal C}_2$ be the collections of all the corresponding values,

1404: indexed by ${\mathbb Z}^2$. We take $c_2(x):=$ black for all $x\in int(\Delta H)$.

1405: 

1406: We shall perform $L$ algorithms, where the $k$-th algorithm corresponds to searching for black paths in $B_k$.

1407: We start with $k=1$.

1408: Let ${\mathbb G}_k$ be the collection of all the (self-avoiding) paths in $B_k\cap AJ$, leading from $\Pi _v

1409: ^b$ to $\Pi _u^{w*}$.

1410: We equip ${\mathbb G}_k$ with an arbitrary deterministic ordering. We also order the vertices along each

1411: path, starting from

1412: $\Pi _v ^b$, going towards $\Pi _u^{w*}$. As before, the $j$-th vertex in the $i$-th path in $B_k\cap AJ$ is

1413: denoted by $_k v _i ^j$.

1414: The algorithm which generates $\xi $ and $\xi ^{(c_2)}$ is the same as in Part 2, using the ``auxiliary''

1415: variable $c$ and the index variables $i$ and $j$.

1416: 

1417: \begin{enumerate}

1418: \item $i:=1,j:=1$.

1419: \item $c:=c_1(_k v _i ^j)$.

1420: \item

1421: \begin{itemize}

1422: \item If $c=$ black, $j:=j+1$.

1423: \item If $c=$ white, $i:=i+1$ and $j:=1$. Stop if $i>|{\mathbb G}_k|$.

1424: \item If $c=$ undefined, with probability $r$, let $c:=$ black, and with probability $1-r$, let

1425: $c:=$ white. Then set $c_1(x):=c$ for all $x\in C^p _{_k v _i ^j}(\eta )$ (i.e., for all $x$ in

1426: the same $\eta$ $p$-cluster of the current vertex), and $c_2(x):=c$ for all

1427: $x\in C^p _{_k v _i ^j}(\eta ^{(c_2)})$.

1428: \end{itemize}

1429: \item Stop if ${\cal C}_1$ contains a black path from $\Pi _v ^b$ to $\Pi _u^{w*}$ in $B_k\cap AJ$,

1430: otherwise go back to 2.

1431: \end{enumerate}

1432: 

1433: When the algorithm terminates, we increase $k$ by one, and if $k\leq L$, we re-run the

1434: algorithm with the new value of $k$.

1435: After the last algorithm stops, we set $\xi(x)=c_1(x)$ for all $x$ such that $c_1(x)$ is not undefined,

1436: and $\xi^{(c_2)}(x)=c_2(x)$ for all $x$ such that $c_2(x)$ is not undefined.

1437: We then assign colour black with probability $r$ and white with probability $1-r$ to the $p$-clusters

1438: in $\eta$ and in $\eta^{(c)}$ that have not been assigned a colour yet, independently of each other.

1439: Here, we make three important remarks.

1440: 

1441: (1) Due to the coupling, the bond configurations $\eta $ and $\eta ^{(c_2)}$ are the same

1442: in $ext(\Delta _1)\cap ext(\Delta _2)\cap ext(\Delta H)$.

1443: Therefore,

1444: if for all $k=1,\ldots ,L$, there is no vertex in $B_k\cap AJ\cap ext(\Delta H)$ with

1445: a dependence range larger than $\frac{m_k}{100}$ in $\eta $, then the same is true in $\eta ^{(c_2)}$ as

1446: well.

1447: By re-writing

1448: (\ref{midiff}) as

1449: $\frac{m_{k+1}}{2}>\frac{3m_k}{2}+\frac{m_k}{100}+\frac{m_{k+1}}{100}$,

1450: we see that the

1451: $\frac{m_k}{100}$-neighbourhoods of the annuli $B_k$ are disjoint.

1452: Hence, if for all $k=1,\ldots ,L$, there is no vertex in $B_k$ with

1453: a dependence range larger than $\frac{m_k}{100}$ in $\eta $, then any point gets

1454: $c_1$ or $c_2$ values by at most one of the algorithms.

1455: 

1456: (2) Similarly to Part 2, the configurations $\xi$ and $\xi^{(c)}$ generated in

1457: the way described above are distributed according to the correct distributions, ${\mathbb P}_r$

1458: and ${\mathbb P}_r^{(c_2)}$ respectively. Note that assigning black to $\xi ^{(c_2)}(x)$ for all

1459: $x\in int(\Delta H)$ is justified since, by the definition of $H$, every such $x$

1460: is connected to $\Pi _v^b$ by an open path in $\eta ^{(c_2)}$.

1461: 

1462: (3) For any $k$, if there is no vertex in $B_k\cap AJ$ with a dependence range larger than

1463: $\frac{m_k}{100}$, then

1464: before the very last step of the $k$-th algorithmic procedure, whenever $c_1(x)$ is black for $x$

1465: in $ext(\Delta_1)\cap ext(\Delta _2)\cap B_k$, $c_2(x)$ is also black for that same $x$.

1466: 

1467: To see this, we need to notice that due to the coupling between $\eta$ and $\eta^{(c_2)}$,

1468: the $p$-clusters of a vertex $x\in ext(\Delta_1)\cap ext(\Delta _2)\cap B_k$ in $\eta$ and in $\eta^{(c_2)}$

1469: may differ in the following four cases:

1470: \begin{itemize}

1471: \item $x\in int(\Delta H)$,

1472: \item $C_x ^p (\eta )$ ``crosses'' $\Delta H$,

1473: \item $C_x ^p (\eta )$ ``crosses'' $\Delta_2$,

1474: \item $C_x ^p (\eta )$ ``crosses'' $\Delta_1$.

1475: \end{itemize}

1476: The difference in the first case is unimportant since we have $c_2(x)=$black for all $x\in int(\Delta H)$.

1477: Recall that $\Delta H$ is a closed barrier both in $\eta$ and in $\eta ^{(c_2)}$; hence the

1478: second case never happens. The $k$-th algorithm assigns values to vertices in $B_k\cap AJ$ only.

1479: If there is no vertex in $B_k\cap AJ$ with a dependence range larger than $\frac{m_k}{100}$,

1480: then the third case does not happen either: $\Delta H$ prevents $C_x ^p(\eta )$ for $x\in B_k\cap AJ$

1481: from intersecting $\Delta _2$.

1482: The fourth case is handled exactly the same way as in Part 2: if such a $p$-cluster is coloured white,

1483: it makes ${\cal C}_1$ ``more white'' than ${\cal C}_2$; if it is black, the algorithm has found

1484: an appropriate black path in ${\cal C}_1$ and therefore terminates.

1485: 

1486: This shows that, for every $k=1,\ldots,L$, if there are no large $\eta $ $p$-clusters in $B_k$,

1487: the presence of a black path in ${\cal C}_1$ from $\Pi _v ^b$ to $\Pi _u^{w*}$ in $B_k\cap AJ$

1488: implies that there is a black path in ${\cal C}_2$ from $\Pi _v ^b$ to $\Pi _u^{w*}$.

1489: Remark (1) above shows that this black path in ${\cal C}_2$ is indeed contained in

1490: $B_k[\frac{m_k}{100}]\cap AJ$.

1491: 

1492: This implies that if we let $T_{B_i}:=\bigcap _{v\in

1493: B_i}\{\mathcal{D}(v)<\frac{m_i}{100}\},$ $B(B_i):=\{$there is a

1494: black circuit in $B_i$ surrounding $x_0\}$ for $i=1,\ldots ,L$, and

1495: $T:=\bigcap _{i=1} ^LT_{B_i}$, we obtain a.s.

1496: \begin{displaymath}

1497: \mathbb{P}_r^{(c_2)}(Q)\geq \mathbb{P}_{p,r}\ (\textrm{$B(B_i)$

1498: holds for at least $\frac{32}{(r_2-r_1) \alpha ^{63} \gamma }+1$

1499: indices}|T)\nu _p(T).

1500: \end{displaymath}

1501: 

1502: The second factor is very close to one as

1503: \begin{eqnarray}

1504: 1-\nu _p(T) & = & \nu _p(\bigcup _{i=1} ^L T_{B_i}^c) \nonumber \\

1505: & \leq & \sum \limits _{i=1} ^L \sum \limits _{v\in B_i}\nu _p(\mathcal{D}(v)\geq \frac{m_i}{100}) \nonumber \\

1506: & \leq & \sum \limits _{i=1} ^L (8m_i^2+8m_i)\nu _{p}(\mathcal{D}(0)\geq \frac{m_i}{100}) \nonumber \\

1507: & \leq & L\cdot \frac{\alpha ^4}{2L} = \alpha ^4/2 \label{Tbig},

1508: \end{eqnarray}

1509: where we used translation invariance, the monotonicity of the function

1510: $f(x)=(8x^2+8x)e^{-\psi(p)\frac{x}{100}}$ above

1511: $x=\frac{600}{\psi (p)}$, and inequalities (\ref{m1nagy}) and (\ref{sokkicsi}).

1512: Note that, conditioned on $T$, the event $B(B_i)$ depends on the

1513: $\frac{m_i}{100}$-neighbourhood of $B_i$ only. We know the

1514: $\frac{m_i}{100}$-neighbourhoods of the annuli $B_i$ are disjoint.

1515: Therefore, the events $B(B_i)$ ($i=1,\ldots ,L$) are conditionally

1516: independent, conditioned on $T$.

1517: We also have, for $i=1,\ldots ,L$,

1518: $$

1519: \mathbb{P}_{p,r}(B(B_i)|T)\geq \mathbb{P}_{p,r}(B(B_i))-\nu_{p}(T^c)\geq \alpha ^4-\alpha ^4/2,

1520: $$

1521: due to (\ref{anncross}) and (\ref{Tbig}). Hence, by the choice of

1522: $L$ before inequality (\ref{sokkicsi}),

1523: \begin{displaymath}

1524: \mathbb{P}_{p,r}\ (\textrm{$B(B_i)$ holds for at least

1525: $\frac{32}{(r_2-r_1) \alpha ^{63} \gamma }+1$ indices}|T)\geq

1526: 1/2.

1527: \end{displaymath}

1528: This shows that

1529: $$

1530: \mathbb{P}_r^{(c_2)}(Q)\geq (1-\frac{\alpha ^4}{2})\frac{1}{2}\geq \frac{1}{4},

1531: $$

1532: proving (\ref{qwerty}).

1533: 

1534: Note that whenever $Q_i$ happens, there is a pivotal (for the event $H_{N,8N}^{w*}$)

1535: $p$-cluster in or close to $B_i$. Moreover, for any $i\neq j$, the events $Q_i$ and $Q_j$

1536: give rise to different pivotal clusters. Therefore, conditioning on having reached Part 3,

1537: the conditional probability of the event $E_3:=\{$there are at least

1538: $\frac{32}{(r_2-r_1) \alpha ^{63}\gamma }+1$ pivotal clusters for

1539: $H_{N,8N}^{w*}\}$ is at least the conditional probability of $Q$,

1540: which is at least $\frac{1}{4}$, as we have just concluded.

1541: 

1542: Since it is easy to see that for any $r\in \left[ r_1,r_2 \right]$ the $\mathbb{P}_{r}$-probability of

1543: reaching Part 3 is at least

1544: $\gamma \cdot (1-\frac{\alpha ^4}{2L})\cdot \frac{\alpha ^{63}}{2}\cdot (1-\frac{\alpha ^4}{2L})\geq \frac{\gamma \alpha

1545: ^{63}}{8}$, and we know that

1546: $n\left( H_{N,8N}^{w*}\right)$ is a nonnegative random variable, we have for any $r\in \left[ r_1,r_2

1547: \right]$,

1548: 

1549: \begin{displaymath}

1550: \mathbb{E}_{r}\left( n\left( H_{N,8N}^{w*}\right) \right) \geq \

1551: \left( \frac{32}{(r_2-r_1) \alpha ^{63} \gamma }+1\right) \cdot \frac{1}{4} \cdot \frac{\gamma \alpha

1552: ^{63}}{8} >\frac{1}{(r_2-r_1) },

1553: \end{displaymath}

1554: finishing the proof of inequality (\ref{pivotal}), and completing the proof of Theorem \ref{main}.

1555: \hfill $\Box $

1556: 

1557: 

1558: \section{Proofs of the remaining results}

1559: \label{consequences}

1560: 

1561: For the proof of Theorem \ref{criticality}, we need the following result of Russo~\cite{russo1}.

1562: Let $\mu $ be a probability measure that assigns colours black or white to the

1563: vertices of $\mathbb{Z}^2$. Let $P^b _{\infty }(\mu )$ (resp. $P^{w*} _{\infty }(\mu )$) denote the

1564: probability that the black

1565: cluster (resp. white $*$-cluster) of the origin is infinite.

1566: Let $S^b(\mu )$ denote the mean size of the black cluster of the origin.

1567: 

1568: \begin{Theorem}\emph{(\cite{russo1})}

1569: \label{meansize}

1570: If $\mu $ is translation invariant and $S^b(\mu )<\infty $, then $P^{w*} _{\infty

1571: }(\mu )>0$.

1572: \end{Theorem}

1573: 

1574: \noindent \textbf{Proof of Theorem \ref{criticality}.} First, we

1575: shall prove criticality when $p<1/2$, $r=r_c(p)$. Our argument

1576: follows the proof of Proposition 1 in \cite{Russocrpr}. Fix $p<1/2$.

1577: Take $\varepsilon >0$ as in Lemma~\ref{finitesizecrit}.

1578: By Lemma~\ref{lexpdecay} and the monotonicity of the function

1579: $f(x)=(x+1)(3x+1)e^{-\psi (p)\frac{x}{3}}$ for $x$ large enough,

1580: there exists $N_0=N_0(p)$ such that, for all $n\geq N_0$,

1581: \begin{equation}\label{nsmalldep}

1582: (n+1)(3n+1)\nu _p(\mathcal{D}(0) \geq \frac{n}{3}) \leq \varepsilon.

1583: \end{equation}

1584: Since $\Theta (p,r)=0$ for all $r<r_c(p)$, Lemma~\ref{finitesizecrit}

1585: and (\ref{nsmalldep}) imply that

1586: $$\mathbb{P}_{p,r}(V_{n,3n} ^b)\leq 1-\varepsilon$$

1587: for all $r<r_c(p),n\geq N_0$.

1588: 

1589: We claim that for any $n$, the function $\mathbb{P}_{p,r}(V_{n,3n}

1590: ^b)$ is continuous in $r$. To see this, notice that the occurrence

1591: of $V_{n,3n}^b$ is completely determined by a partitioning of the

1592: vertices in $S_{n,3n}$ in $p$-clusters, and the colours assigned to

1593: these clusters. Let us denote by $\mathcal{P}_{S}$

1594: the set of partitions of the

1595: vertices in $S_{n,3n}$ which are compatible with a bond configuration,

1596: and the (random)

1597: partition determined by the initial bond percolation by

1598: $\mathcal{G}_{S}$. Fix an arbitrary partition $g_{S}\in

1599: \mathcal{P}_{S}$. Since the colours are assigned independently to

1600: the $p$-clusters determined by $g_{S}$, it is easy to see that

1601: $\mathbb{P}_{p,r}(V_{n,3n} ^b|\mathcal{G}_{S}=g_{S})$ is a

1602: polynomial function of $r$, hence continuous in $r$.

1603: This implies that the (finite) linear combination

1604: $$\mathbb{P}_{p,r}(V_{n,3n} ^b)=\sum \limits _{g_{S}\in \mathcal{P}_{S}}

1605: \nu _p(\mathcal{G}_{S}=g_{S})\ \mathbb{P}_{p,r}(V_{n,3n}

1606: ^b|\mathcal{G}_{S}=g_{S})$$ is indeed continuous in $r$.

1607: 

1608: This shows that for any $n\geq N_0$, if we let $r\to r_c$, we obtain

1609: $\mathbb{P}_{p,r_c}(V_{n,3n} ^b)\leq 1-\varepsilon $. Therefore,

1610: $$\limsup \limits _{n\to \infty }\mathbb{P}_{p,r_c}(V_{n,3n} ^b)<1,$$

1611: which, by Lemma \ref{lessoreq}, implies $\Theta (p,r_c)=0$,

1612: providing the first condition of criticality.

1613: 

1614: The relation $\Theta ^*(p,r_c^*)=0$ can be proved analogously.

1615: Hence, as $r_c^*=1-r_c,$ we obtain that the

1616: $\mathbb{P}_{p,r_c}$-probability of the origin being in an infinite

1617: white $*$-cluster is 0. Therefore, applying Theorem \ref{meansize}

1618: to the measure $\mathbb{P}_{p,r_c}$ yields that the mean size of

1619: the black cluster of the origin is infinite, concluding the proof

1620: of criticality for $p<1/2$, $r=r_c(p)$.

1621: 

1622: The fact that there is no infinite black cluster or white

1623: $*$-cluster at $p=1/2$, $r\in (0,1)$,

1624: is a straightforward

1625: consequence of the fact that $\nu_{1/2}$-almost every

1626: $p$-configuration contains infinitely many disjoint open

1627: circuits surrounding the origin. These circuits are coloured

1628: independently, preventing the possibility of black percolation

1629: or white $*$-percolation.

1630: (This idea has been described in \cite{DaC} already

1631: to show that there is no percolation of either colour at $p=1/2, r=1/2$.)

1632: The infinite mean cluster size follows

1633: then from Theorem~\ref{meansize}, as before.

1634: 

1635: The supercritical case $p>1/2$ is obvious: the probability that the

1636: origin is in an infinite $p$-cluster is positive in that case, and

1637: so is the probability that the colour assigned to that cluster is

1638: black for any $r>0$. \hfill $\Box$\\

1639: 

1640: \begin{Remark}

1641: {\rm The proof of $r_c+r_c^*=1$ for $p<1/2$ uses the FKG inequality, exponential

1642: decay of correlations and duality.

1643: It is easy to see that polynomial decay of correlations of degree strictly

1644: greater than 2 would be enough for the proof.

1645: The fact that there is no infinite black cluster at $p=1/2$, $r\in(0,1)$,

1646: (Theorem \ref{criticality}) even though duality and the FKG inequality hold

1647: in that case, shows that at $p=1/2$, for any $c>2$ and $N\in \mathbb{N}$,

1648: there exists $n>N$ such that

1649: \begin{displaymath}

1650: \nu _{1/2}(\mathcal{D}(0)\geq n)\geq \frac{1}{n^c},

1651: \end{displaymath}

1652: i.e., in critical bond percolation on the square lattice, the probability that

1653: the origin is connected to $\partial B_n$ by an open path is at least $n^{-c}$.

1654: }

1655: \end{Remark}

1656: 

1657: 

1658: \noindent \textbf{Proof of Corollary \ref{phase-diagram}.}

1659: To prove Corollary \ref{phase-diagram}, one needs to put together the results in Theorems

1660: \ref{duality}--\ref{criticality}. Strictly speaking,

1661: the following three statements need additional clarification: for $p<1/2$, we have

1662: 

1663: 

1664: (i) $r_c(p) \in [1/2, 1)$,

1665: 

1666: (ii) $\Theta ^*(p,1-r_c(p))=0$ and the mean size of the white $*$-cluster of the origin is infinite, and

1667: 

1668: (iii) If $r>r_c(p)$, the size of the white $*$-cluster has an exponentially decaying tail.

1669: 

1670: \medskip

1671: Now $r_c(p) <1$ follows from Theorem 2.6 in

1672: \cite{DaC}.

1673: The other bound $r_c(p) \geq 1/2$ is an easy consequence of $r_c(p) + r_c^*(p) =1$, since

1674: $r_c(p) \geq r_c^*(p)$.

1675: We have seen the first half of (ii), i.e. $\Theta ^*(p,r_c^*(p))=0$, in the proof of Theorem \ref{criticality}.

1676: We also know that $\Theta(p,r_c(p))=0$, which implies,

1677: according to Theorem \ref{meansize}, that the mean size of the white $*$-cluster of the origin is infinite.

1678: Statement (iii) can be proved the same way as Theorem \ref{exponential-decay}.

1679: \hfill $\Box $\\

1680: 

1681: The proof of Corollary \ref{continuity} uses the methods of Russo \cite{russo1}, and van den Berg and

1682: Keane \cite{RobKeane},

1683: based on the following lemma, which may be interesting in itself.

1684: 

1685: \begin{Lemma}

1686: At $p<1/2$, $r>r_c(p)$, the number of infinite black clusters is $\mathbb{P}_{p,r}$-a.s.\ equal to 1.

1687: \end{Lemma}

1688: 

1689: \noindent \textbf{Proof.}

1690: For $r_c<r<1$, similarly to the proof of Theorem \ref{groreq},

1691: conditions (1)-(4) of Theorem \ref{GanKeR} clearly hold for the measure $\mathbb{P}_{p,r}$.

1692: Theorem~\ref{GanKeR} states that under these conditions, the number of infinite black clusters is 1.

1693: The case $r=1$ is obvious.

1694: \hfill $\Box $\\

1695: 

1696: \noindent \textbf{Proof of Corollary \ref{continuity}.}

1697: We fix $p<1/2$, and write $\Theta(r|\eta)$ for the conditional probability that the

1698: cluster of the origin is infinite, given that the bond configuration is $\eta$. The above

1699: mentioned classical arguments

1700: and Theorem \ref{criticality} give that $\Theta(r|\eta)$ is for almost all $\eta$ a continuous function

1701: in $r$.

1702: 

1703: Now fix $\varepsilon >0$ and $r \in [0,1]$. For almost every $\eta$, there exists a maximal $\delta(\eta)$ such that

1704: if $|r'-r| \leq \delta(\eta)$, then $|\Theta(r|\eta)-\Theta(r'|\eta)| < \varepsilon$. Now choose $\bar{\delta}>0$

1705: so small that $\nu_p(\eta; \delta(\eta) < \bar{\delta}) < \varepsilon$ and denote the set $\{\eta; \delta(\eta) \geq

1706: \bar{\delta}\}$ by $A$. Since $\Theta(p,r)=\int \Theta(r|\eta)d\nu_p(\eta)$ we then find that for $r'$ such that

1707: $|r-r'| < \bar{\delta}$, we have

1708: \begin{eqnarray*}

1709: |\Theta(p,r)-\Theta(p,r')| & \leq &  \int_A |\Theta(r|\eta)-\Theta(r'|\eta)| d\nu_p(\eta)

1710: + \\

1711: & & + \int_{A^c} |\Theta(r|\eta)-\Theta(r'|\eta)| d\nu_p(\eta)\\

1712: &\leq& \varepsilon \nu_p(A) + 2\nu_p(A^c)\\

1713: &\leq & \varepsilon + 2\varepsilon,

1714: \end{eqnarray*}

1715: proving the result.

1716: \hfill $\Box $

1717: 

1718: 

1719: \section{The DaC model on the triangular lattice $\mathbb{T}$}

1720: \label{triang}

1721: On the square lattice, the relationship $r_c(p)+r_c^*(p)=1$ does not determine

1722: the critical value $r_c(p)$. However, on the triangular lattice, percolation is

1723: self-dual (i.e., $*$-paths are the same as ordinary paths), so that the same

1724: relationship immediately implies $r_c^*(p) =r_c(p)=1/2$.

1725: In this section, we elaborate a bit on the proof of $r_c(p)+r_c^*(p)=1$ for

1726: $p<p_c(\mathbb{T})$ on the triangular lattice. In this case, the version of the

1727: RSW-type theorem of Bollob\'as and Riordan~\cite{Voronoi} suffices, and we do not

1728: need to use the improvement in~\cite{BBB}.

1729: 

1730: We embed the triangular lattice $\mathbb T$ in ${\mathbb R}^2$ so that its vertices $\mathcal{V}(\mathbb T)$

1731: are the intersections of the lines $y=-\sqrt{3}\hspace{0.05cm} x+\sqrt{3}\hspace{0.05cm} k$ and

1732: $y=\frac{\sqrt{3}}{2}\hspace{0.05cm} l$ for $k,l \in \mathbb{Z}$, and denote the elements of

1733: $\mathcal{V}(\mathbb T)$ by $(k,l)$. For example, $\left( 0,0\right)$ refers to the intersection of

1734: $y=-\sqrt{3}\hspace{0.05cm}x$ and $y=0$.

1735: The edges are given by

1736: $\mathcal{E}\left( \mathbb{T} \right) := \{ (a,b): \hspace{0.2cm} a,b\in \mathcal{V}\left( \mathbb{T}\right) ,|\hspace{0.05cm}

1737: a-b\hspace{0.05cm}|=1\} $, where $|\cdot |$ denotes the Euclidean norm (see Figure \ref{rec}).

1738: We define and denote paths, circuits, horizontal and vertical crossings exactly as before.

1739: Note that $S_{m,n}$ corresponds to a parallelogram in $\mathbb{R}^2$ of side lengths $m$ and $n$,

1740: as in the example in Figure~\ref{rec}.

1741: 

1742: Given the equivalence between crossings and $*$-crossings, we will drop the $*$ from our

1743: notation in this section. We note that the definitions and all the preliminary results of

1744: Section~\ref{prelim} still apply, modulo the reinterpretation of $*$-crossings as ordinary

1745: crossings and the different critical value.

1746: This observation will be implicitly understood in the rest of the section and we will use

1747: the results of Section~\ref{prelim-results} without further comments.

1748: In this section, $p_c$ denotes $p_c(\mathbb{T})$, the critical value for bond percolation on $\mathbb{T}$.

1749: 

1750: \begin{figure}[h]

1751: \centering

1752: \includegraphics[scale=0.37, trim= 0mm 0mm 0mm 0mm, clip ]{abra3.EPS}

1753: \caption{Portion of the triangular lattice $\mathbb{T}$. The heavy segments are the sides of

1754: the parallelogram $S_{2,3}$.}

1755: \label{rec}

1756: \end{figure}

1757: 

1758: The inequality $r_c(p) \geq 1/2$ can be proved by standard methods.

1759: Similar (however somewhat simpler) considerations to those in the proof of Theorem~\ref{main}

1760: lead to $r_c(p)\leq 1/2$ as follows.

1761: It is easy to see that for any rhombus $S$, the probability of having a black vertical crossing

1762: is exactly the same as the probability of having a black horizontal crossing in $S$.

1763: This observation, together with Lemma~\ref{selfdl} and symmetry of black and white at $r=1/2$

1764: implies the following result.

1765: 

1766: \begin{Lemma}

1767: \label{rhcross}

1768: For any $p\in [0,1]$ and any $n\in \mathbb{N}$,

1769: \begin{displaymath}

1770: \mathbb{P}_{p,1/2}(V_{n,n}^b)=1/2.

1771: \end{displaymath}

1772: \end{Lemma}

1773: 

1774: This lemma allows us to use the RSW type theorem of Bollob\'as and Riordan~\cite{Voronoi},

1775: which states that

1776: $\liminf \limits _{n\to \infty }\mathbb{P}_{p,1/2}(V_{n,n}^b)>0$ implies

1777: $\limsup \limits _{n\to \infty }\mathbb{P}_{p,1/2}(V_{n,\rho n}^b)>0$

1778: for any $\rho >0$, to obtain the following result.

1779: 

1780: \begin{Lemma}

1781: \label{poscross}

1782: For any $p<p_c$, we have

1783: \begin{equation}

1784: \limsup \limits _{n\to \infty }\mathbb{P}_{p,1/2}(V_{n,3n}^b)>0.

1785: \end{equation}

1786: \end{Lemma}

1787: 

1788: Next we show that certain parallelograms have high crossing probabilities.

1789: 

1790: \begin{Theorem}

1791: \label{maintri}

1792: For all $p<p_c$, for any $\varepsilon >0$, \hspace{0.05cm}

1793: $$\limsup \limits _{n\to \infty }\mathbb{P}_{p,1/2+\varepsilon }(V_{n,3n}^b)= 1.$$

1794: \end{Theorem}

1795: 

1796: \noindent \textbf{Proof sketch. }We assume that there exists a $p<p_c$ and an $\varepsilon >0$

1797: such that $\limsup \limits _{n\to \infty }\mathbb{P}_{p,1/2+\varepsilon }(V_{n,3n}^b)<1$.

1798: We denote the measure $\mathbb{P}_{p,r}(\cdot )$ by $\mathbb{P}_{r}(\cdot )$.

1799: The assumption above and Lemma \ref{selfdl} gives $\liminf \limits _{n\to \infty

1800: }\mathbb{P}_{1/2+\varepsilon }(H_{n,3n}^{w})>0$,

1801: which, together with monotonicity, shows that in the interval $r\in [1/2,1/2+\varepsilon ]$,

1802: whenever $n$ is large enough,

1803: the $\mathbb{P}_r$-probability of $H_{n,3n}^{w}$ is bounded away from 0.

1804: 

1805: On the other hand, Lemma \ref{poscross} gives $\limsup \limits _{n\to \infty

1806: }\mathbb{P}_{1/2}(V_{n,3n}^b)>0$.

1807: Therefore, there exists a sequence of side lengths $n_k\to \infty$ along which

1808: $\mathbb{P}_{1/2}(V_{n_k,3n_k}^b)$ is also bounded away from 0, which, by monotonicity,

1809: gives the same lower bound in the

1810: whole interval $r\in [1/2,1/2+\varepsilon ]$ for $\mathbb{P}_{r}(V_{n_k,3n_k}^b)$.

1811: 

1812: A careful reading of the proof of Theorem \ref{main} shows that these lower bounds

1813: are enough to determine the parameters $L, m_1,\ldots ,m_L, N$ of the construction described

1814: in the proof of inequality (\ref{pivotal}), which provides a uniform lower bound on the number

1815: of pivotal $p$-clusters for the event $H_{N,8N} ^{w}$ in the interval $r\in [1/2,1/2+\varepsilon ]$,

1816: leading to a contradiction.

1817: \hfill $\Box $\\

1818: 

1819: Theorem \ref{maintri} together with Lemma \ref{lessoreq} implies

1820: $r_c(p)\leq 1/2$, establishing the equality $r_c(p)=1/2$ and completing the proof of

1821: Theorem~\ref{critical-point}.

1822: 

1823: We conclude this section with the proof of Proposition~\ref{universality}. \\

1824: 

1825: \noindent \textbf{Proof of Proposition~\ref{universality}.}

1826: Let us fix $q \geq 2$ and $p<p_c(q)$ and denote the corresponding probability measure

1827: by $\mu_{p,q,r}$.

1828: One can check that conditions (1)--(3) of Theorem~\ref{GanKeR} apply to $\mu_{p,q,1/2}$:

1829: condition (1) is obvious, condition (2) can be found, for example, in~\cite{grimmett2},

1830: condition (3) is proved in~\cite{Haggstrom}.

1831: If we now assume that for $r=1/2$ there exists an infinite black cluster with positive

1832: probability (meaning that condition (4) is also satisfied by $\mu_{p,q,1/2}$), colour

1833: symmetry implies the existence of an infinite white cluster with positive probability,

1834: leading to a contradiction with Theorem~\ref{GanKeR}. We then conclude that there

1835: exists a.s no infinite black cluster at $r=1/2$ and, by colour symmetry again, no

1836: infinite white cluster. Since $\mu_{p,q,1/2}$ is clearly a translation-invariant

1837: measure, Theorem~\ref{meansize} implies infinite mean size for the black $r$-cluster

1838: of the origin.

1839: \hfill $\Box $\\

1840: 

1841: \medskip\noindent

1842: {\bf Acknowledgements} We would like to thank Rob van den Berg for useful and stimulating discussions.

1843: F.C. thanks Reda J\"urg Messikh and Akira Sakai for interesting discussions at an early stage of this work.

1844: 

1845: 

1846: \begin{thebibliography}{99}

1847: 

1848: \bibitem{ab} M.~Aizenman, D.J.~Barsky, Sharpness of the phase transition in percolation models,

1849: \emph{Comm.~Math.~Phys.} {\bf 86}, 1--48 (1987).

1850: 

1851: \bibitem{Robnew}

1852: J. van den Berg,

1853: Approximate zero-one laws and sharpness of the percolation transition in a

1854: class of models including 2D Ising percolation,

1855: preprint available from arXiv:math.PR/0707.2077v1 (2007).

1856: 

1857: \bibitem{BBB}

1858: J. van den Berg, R. Brouwer, B. V\'agv\"olgyi,

1859: Continuity for self-destructive percolation in the plane,

1860: preprint available from arXiv:math.PR/0603223 (2006).

1861: 

1862: \bibitem{RobKeane}

1863: J. van den Berg, M. Keane,

1864: On the continuity of the percolation probability function, \textit{Particle Systems, Random Media and Large

1865: Deviations} (R.T. Durrett, ed.),

1866: Contemporary Mathematics Series \textbf{26}, AMS, Providence, R. I., 61--65 (1984).

1867: 

1868: \bibitem{BH}

1869: S.R. Broadbent, J.M. Hammersley, Percolation processes. I. Crystals and mazes, \textit{Proc. Cambridge

1870: Philos. Soc.}

1871: {\bf 53}, 629--641 (1957).

1872: 

1873: \bibitem{Voronoi}

1874: B. Bollob\'as, O. Riordan,

1875: The critical probability for random Voronoi percolation in the plane is 1/2,

1876: \textit{Probability Theory and Related Fields}

1877: {\bf 136}, 417--468 (2006).

1878: 

1879: \bibitem{shpthr}

1880: B. Bollob\'as, O. Riordan,

1881: Sharp thresholds and percolation in the plane,

1882: preprint available from arXiv:math.PR/0412510 (2004).

1883: 

1884: \bibitem{cn1} F.~Camia, C.M.~Newman,

1885: Continuum Nonsimple Loops and 2D Critical Percolation,

1886: \textit{J.~Stat.~Phys.} {\bf 116}, 157-173 (2004).

1887: 

1888: \bibitem{cn2} F.~Camia, C.M.~Newman,

1889: Two-Dimensional Critical Percolation: the Full Scaling Limit,

1890: \textit{Comm.~Math.~Phys.} {\bf 268}, 1-38 (2006).

1891: 

1892: \bibitem{cardy} J.L.~Cardy, Critical percolation in finite geometries,

1893: \textit{J.~Phys.~A} {\bf 25}, L201-L206 (1992).

1894: 

1895: \bibitem{MM}

1896: M. Franceschetti, R. Meester,

1897: \textit{Random networks for communication}, Cambridge University Press (2007).

1898: 

1899: \bibitem{GKR}

1900: A. Gandolfi, M. Keane, L. Russo,

1901: On the uniqueness of the infinite occupied cluster in dependent two-dimensional site percolation,

1902: \textit{The Annals of Probability} {\bf 16}, 1147--1157 (1988).

1903: 

1904: \bibitem{Grimmett} G. Grimmett,

1905: \textit{Percolation} (2nd ed.), Springer (1999).

1906: 

1907: \bibitem{grimmett2} G. Grimmett, \textit{The random-cluster model},

1908: Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences] {\bf 333},

1909: Springer-Verlag, Berlin (2006).

1910: 

1911: \bibitem{Haggstrom} O. H\"aggstr\"om, Positive correlations in the fuzzy Potts model,

1912: \textit{The Annals of Applied Probability} {\bf 9}, 1149--1159 (1999).

1913: 

1914: \bibitem{DaC}

1915: O. H\"aggstr\"om,

1916: Coloring percolation clusters at random,

1917: \textit{Stochastic Processes and their Applications} {\bf 96}, 213--242 (2001).

1918: 

1919: \bibitem{Kesten} H. Kesten, The critical probability of bond percolation on the square lattice equals

1920: $1/2$,

1921: \textit{Comm. Math. Phys.} {\bf 74}, 41--59 (1980).

1922: 

1923: \bibitem{LigSchSt}

1924: T.M. Liggett, R.H. Schonmann and A.M. Stacey,

1925: Domination by product measures,

1926: \textit{Annals of probability} {\bf 25}, 71--95 (1997).

1927: 

1928: \bibitem{menshikov} M.V.~Menshikov, Coincidence of critical points in percolation problems,

1929: \emph{Soviet Mathematics Doklady} {\bf 33}, 368--370 (1986).

1930: 

1931: \bibitem{russo1} L.~Russo, A note on percolation, \textit{Z. Wahrscheinlichkeitstheorie und Verw. Gebiete}

1932: {\bf 43}, 39--48 (1978).

1933: 

1934: \bibitem{Russocrpr} L. Russo, On the critical percolation probabilities, \textit{Z. Wahrsch. Verw. Gebiete}

1935: {\bf 56}, 229--237 (1981).

1936: 

1937: \bibitem{smirnov} S.~Smirnov, Critical

1938: percolation in the plane: Conformal invariance,

1939: Cardy's formula, scaling limits,

1940: \textit{C.~R.~Acad.~Sci.~Paris} {\bf 333}, 239--244 (2001).

1941: 

1942: \bibitem{se1} M.F. Sykes, J.W. Essam, Some exact critical percolation probabilities for site and bond

1943: problems in two dimensions, \textit{Phys. Rev. Lett.} {\bf 10}, 3--4 (1963).

1944: 

1945: \bibitem{se2} M.F. Sykes, J.W. Essam, Exact critical percolation probabilities for site and bond

1946: problems in two dimensions, \textit{J. Math. Phys.} {\bf 5}, 1117--1127 (1964).

1947: 

1948: \bibitem{wierman} J.C. Wierman, Bond percolation on honeycomb and triangular lattices,

1949: \textit{Adv. in Appl. Probab.} {\bf 13}, 298--313 (1981).

1950: 

1951: \end{thebibliography}

1952: 

1953: \end{document}