1:
2: \documentclass[a4paper,11pt]{article}
3:
4:
5: %\newcommand{\wcon}{{\stackrel{\lambda}{\leftrightarrow}}}
6: %\newcommand{\wcon}{{\stackrel{\star}{\leftrightarrow}}}
7: \newcommand{\wcon}{\leftrightarrow}
8: \newcommand{\CCC}{{\mathcal C}}
9: \newtheorem {theorem}{Theorem}[section]
10: \newtheorem {prop}[theorem]{Proposition}
11: \newtheorem {lemma}[theorem]{Lemma}
12: \newtheorem {corollary}[theorem]{Corollary}
13: \newcounter{conjecture}\setcounter{conjecture}{1}
14: \newcounter{remark}\setcounter{remark}{1}
15: \newenvironment{conjecture}{\medskip{\bf Conjecture.\ }\em}{\rm}
16: \newenvironment{remark}{\medskip{\bf Remark \theremark.}
17: \addtocounter{remark}{1}}{}
18: \newcommand{\eqnsection}{
19: \renewcommand{\theequation}{\thesection.\arabic{equation}}
20: \makeatletter
21: \csname @addtoreset\endcsname{equation}{section}
22: \makeatother}
23: \newcommand{\req}[1]{(\ref{#1})}
24: %\newtheorem {defin}{D}[]
25: \newtheorem {definition}{Definition}[section]
26:
27: \author{Alan Hammond\\Department of Statistics\\University of California, Berkeley\\alanmh@stat.berkeley.edu}
28: \title{Percolation and lattice animals: exponent relations, and conditions for $\theta(p_c)=0$}
29:
30: \usepackage{amsfonts}
31: \usepackage[dvips]{graphicx}
32:
33: \newcommand{\dumy}[1]{\sum_{#1}}
34: \newcommand{\sdash}{\sigma_{n,m}}
35: \newcommand{\sigm}{\sigma_{n,m}p^n(1-p)^m}
36: \newcommand{\signt}{\sigma_{n,m}t_n^n(1-t_n)^m}
37: \newcommand{\sigt}{{\Big( \frac{n}{n+m} \Big)}^{n} {\Big( \frac{m}{n+m} \Big)}^{m}}
38: \newcommand{\sigo}{p_c^n(1 - p_c)^m}
39: \newcommand{\reta}{\varsigma}
40: \newcommand{\rrho}{\varrho}
41: \newcommand{\appak}{\lambda}
42: \newcommand{\lapha}{\alpha}
43: \newcommand{\dummmy}[3]{\sum_{#3}}
44: \newcommand{\suf}[2]{\sum_{n}{\sum_{#1}{#2}}}
45: \newcommand{\suh}[3]{\sum_{n}{\sum_{#1}^{#2}{#3}}}
46: \newcommand{\suff}[3]{\sum_{#1}{\sum_{#2}{#3}}}
47: \newcommand{\atwo}{m > \lfloor n \lapha + n^{1/2} \rfloor + 1}
48: \newcommand{\alphigh}{m \in n B ( \lapha , C ( \log{n} / n )^{1/2} ) }
49: \newcommand{\alplow}{m \in n B ( \lapha , n^{-1/2} ) }
50: \newcommand{\alpnhigh}{m \in n B ( \lapha_n , C ( \log{n} / n )^{1/2} ) }
51: \newcommand{\alpnlow}{m \in n B ( \lapha_n , n^{-1/2} ) }
52:
53: \begin{document}
54: \maketitle
55:
56: {\bf Abstract}
57: We examine the percolation model on $\mathbb{Z}^d$ by an approach
58: involving lattice animals and their surface-area-to-volume ratio.
59: For $\beta \in [0,2(d-1))$, let
60: $f(\beta)$ be the asymptotic exponential rate in the number
61: of edges of the number of lattice animals containing the origin which
62: have surface-area-to-volume ratio $\beta$. The function $f$ is bounded
63: above by a function which may be written in an explicit form. For low
64: values of $\beta$ (\mbox{$\beta \leq 1/p_c - 1$}), equality holds, as
65: originally demonstrated by F.Delyon. For higher values ($\beta > 1/p_c
66: - 1$), the inequality is strict.
67:
68: We introduce two critical exponents, one of which describes how
69: quickly $f$ falls away from the explicit form as $\beta$ rises from
70: $1/p_c - 1$, and the second of which describes how large clusters
71: appear in the marginally subcritical regime of the percolation model.
72: We demonstrate that the pair of exponents must satisfy certain
73: inequalities, while other such inequalities yield sufficient
74: conditions for the absence of an infinite cluster at the critical
75: value. The first exponent is related to one of a more conventional nature in
76: the scaling theory of percolation, that of correlation size. In
77: deriving this relation, we find that there are two possible
78: behaviours, depending on the value of the first exponent, for the
79: typical surface-area-to-volume ratio of an unusually large cluster in
80: the marginally subcritical regime.
81:
82: This paper provides an account of the central aspects of the approach,
83: including the proofs of the main results. In the longer report
84: \cite{techrep}, complete proofs of all of the assertions are given.
85:
86: {\bf Keywords} Percolation, lattice animals, critical exponents.
87:
88: \section{Introduction}
89:
90: Percolation on the integer lattice $\mathbb{Z}^d$ is one of the most
91: fundamental and intensively studied models in the rigorous theory of
92: statistical mechanics. Many aspects of the behaviour of the model in
93: the subcritical and supercritical regime have been determined
94: rigorously. The problem of understanding the behaviour of the model at
95: criticality, and interplay between this behaviour and that for
96: parameter values nearby, has been addressed widely by physicists, but
97: the search for proofs of many of their predictions continues. These
98: predictions typically take the form of asserting the value of critical
99: exponents, and thereby describe the power-law decay or explosion of
100: characteristics of the model near criticality.
101:
102: In this paper, we examine the percolation model by an
103: approach involving lattice animals, divided according to their
104: surface-area-to-volume ratio.
105: Throughout, we work with the bond percolation
106: model in $\mathbb{Z}^d$. However, the results apply to the site or
107: bond model on any infinite transitive amenable graph with inessential changes.
108:
109: For any given $p \in (0,1)$, two lattice animals with given size are
110: equally likely to arise as the cluster $C(0)$ containing the origin
111: provided that they have the same surface-area-to-volume ratio.
112: For given $\beta \in (0,\infty)$, there is an exponential growth rate
113: in the number of edges for the number of lattice animals up to translation that have surface-area-to-volume ratio very close to $\beta$. This growth rate $f(\beta)$ may be studied as a function of $\beta$.
114: To illustrate the connection between the percolation model and the combinatorial question of the behaviour of $f$, note that the probability that the cluster containing the origin contains a large number $n$ of edges is given by
115: \begin{displaymath}
116: \mathbb{P}_p(\vert C(0) \vert = n) = \sum_{m}{\sigma_{n,m}p^n(1-p)^m},
117: \end{displaymath}
118: where $\sigma_{n,m}$ is the number of lattice animals that contain the origin, have $n$ edges and $m$ outlying edges. We rewrite the right-hand-side to highlight the role of the surface-area-to-volume ratio, $m/n$:
119: \begin{equation}\label{rrr}
120: \mathbb{P}_p(\vert C(0) \vert = n) = \sum_{m}{(f_n(m/n)p (1-p)^{m/n})^{n}}.
121: \end{equation}
122: Here $f_n(\beta) = (\sigma_{n,\lfloor \beta n \rfloor})^{1/n}$ is a
123: rescaling that anticipates the exponential growth that occurs. We
124: examine thoroughly the link between percolation and combinatorics
125: provided by Equation \ref{rrr}.
126:
127: An overview of the approach is now given, in the form of a
128: description of the organisation of the paper.
129: In Section \ref{secttwo}, we describe the model, and define notations, before
130: stating the combinatorial results that we will use. The proofs are
131: largely omitted, as are a few results in later sections. (We refer the
132: interested reader to the report \cite{techrep}, in which all proofs are given
133: in full, along with some notes on the literature.) The
134: combinatorial results assert the existence of the function $f$ and
135: describe aspects of its behaviour, Theorem \ref{thmtwo} implying that
136: \begin{equation}\label{fgh}
137: \log f(\beta) \leq (\beta + 1)\log(\beta + 1) - \beta \log \beta \
138: \textrm{for $\beta \in (0,2(d-1))$}.
139: \end{equation}
140: F.Delyon \cite{DelyonT} showed that equality holds for $\beta \in (0,1/p_c -
141: 1)$. Theorem \ref{thmtwo} implies that the inequality is strict for higher values of
142: $\beta$. The marked change, as $\beta$ passes through $1/p_c - 1$, in the structure of large
143: lattice animals of surface-area-to-volume ratio $\beta$ is a
144: combinatorial analogue of the phase transition in percolation at criticality.
145: The notion of a collapse transition for animals has been explored in \cite{MR1304212}.
146:
147: In Section \ref{sectthr}, two scaling hypotheses are
148: introduced, each postulating the existence of a critical exponent. One of the
149: exponents, $\reta$, describes how quickly $f(\beta)$ drops away from
150: the explicit form given on the right-hand-side of [\ref{fgh}] as $\beta$ rises above $1/p_c - 1$. The other,
151: $\appak$, describes how rapidly decaying in $n$ is the discrepancy between the
152: critical value and that value on the subcritical interval at which the
153: probability of observing an $n$-edged animal as the cluster to
154: which the origin belongs is maximal. The first main result, Theorem
155: \ref{thmthr}, is then proved: the inequalities $\appak < 1/2$ and $\reta
156: \appak < 1$ cannot both be satisfied, because they imply that the mean
157: cluster size is uniformly bounded on the subcritical interval,
158: contradicting known results.
159:
160: In Section \ref{sectfour}, sufficient conditions for the absence of an
161: infinite cluster at the critical value are proved. Theorem
162: \ref{thmfour} asserts that $\reta < 2$ or $\appak > 1/2$ are two such
163: conditions.
164: Except for some borderline cases, the range of values remaining after
165: Theorems \ref{thmthr} and \ref{thmfour} is specified by $\appak < 1/2$ and $\reta \appak >
166: 1$. In Theorem \ref{thmund}, where we see that in this case, such a sufficient
167: condition may be expressed in terms of the extent to which the asymptotic
168: exponential rate $f(\beta)$ is underestimated by its finite
169: approximants $f_n(\beta)$ for a certain range of values of $\beta$.
170: The extent of underestimation is related to combinatorial exponents
171: such as the entropic exponent (see, for example, \cite{MR95m:82076}).
172:
173: In Section \ref{sectfive}, we relate the value of $\reta$ to an
174: exponent of a more conventional nature in the scaling theory of
175: percolation, that of
176: correlation size (see Theorem \ref{thmfive}).
177: Suppose that we perform an experiment in which the
178: surface-area-to-volume ratio of the cluster to which the origin
179: belongs is observed, conditional on its having a very large number of
180: edges, for a $p$-value slightly below $p_c$.
181: How does the typical measurement, $\beta_p$, in this experiment behave
182: as $p$ tends to $p_c$?
183: The value $\beta_p$ tends to lie somewhere on the interval
184: \mbox{$(1/p_c - 1,1/p - 1)$}.
185: In Theorem \ref{thmsix}, we determine that there are two possible scaling
186: behaviours. The inequality $\reta < 2$ again arises, distinguishing the
187: two possibilities. If $\reta < 2$, then $\beta_p$ scales much closer to
188: $1/p_c - 1$ while if $\reta > 2$, it is found
189: to be closer to $1/p - 1$.
190:
191:
192: \section{Notations and combinatorial results}\label{secttwo}
193:
194: Throughout, we work with the bond percolation model on $\mathbb{Z}^d$,
195: for any given $d \geq 2$.
196: This model has a parameter $p$ lying in the interval $[0,1]$. Nearest
197: neighbour edges of $\mathbb{Z}^d$ are declared to be open with
198: probability $p$, these choices being made independently between
199: distinct edges. For any vertex $x \in \mathbb{Z}^d$, there is a
200: cluster $C(x)$ of edges accessible from $x$, namely the collection of
201: edges that lie in a nearest-neighbour path of open edges one of whose members contains
202: $x$ as an endpoint. The percolation probability $\theta(p)$ as a
203: function of $p$ may then be written \mbox{$\theta(p)= \mathbb{P}(
204: \vert C(0) \vert = \infty)$}. To demonstrate the continuity of
205: $\theta$, it suffices to show that $\theta(p_c) = 0$ (cf \cite{grim}), where $p_c$
206: denotes the critical value, namely the infimum of those values of $p$
207: for which $\theta$ is positive.
208:
209: \begin{definition}
210: A lattice animal is the collection of edges of a finite connected
211: subgraph of $\mathbb{Z}^d$. An edge of $\mathbb{Z}^d$ is said to be
212: outlying to a lattice animal if it is not a member of the animal, and
213: if there is an edge in the animal sharing an endpoint with this
214: edge.
215: We adopt the notations:
216: \begin{itemize}
217: \item for $n,m \in \mathbb{N}$, set $\Gamma_{n,m}$ equal to the collection
218: of lattice animals in $\mathbb{Z}^d$ one of whose edges contains the
219: origin, having $n$ edges, and $m$ outlying edges. Define
220: $\sigma_{n,m} = \vert \Gamma_{n,m} \vert$. The surface-area-to-volume
221: ratio of any animal in $\Gamma_{n,m}$ is said to be $m/n$.
222: \item
223: for each $n \in \mathbb{N}$, define the function $f_n: [0,\infty) \to [0,\infty)$ by
224: \begin{displaymath}
225: f_n(\beta)=(\sigma_{n,\lfloor \beta n \rfloor})^{1/n}
226: \end{displaymath}
227: \end{itemize}
228: \end{definition}
229: On another point of notation, we will sometimes write the index set of
230: a sum in the form $n S$, with $S \subseteq (0,\infty)$, by which is meant
231: $\{m \in \mathbb{N}: m/n \in S \}$. \\
232: We require some results about the asymptotic exponential growth rate
233: of the number of lattice animals as a function of their
234: surface-area-to-volume ratio. The proofs of the theorems stated here
235: are given in \cite{techrep}.
236: \begin{theorem} \label{thmone}
237: ${}$
238: \flushleft
239: \begin{enumerate}
240: \item For $\beta \in [0,\infty) - \{ 2(d-1) \}$, $f(\beta)$ exists,
241: being defined as the limit $\lim_{n \to \infty} f_n(\beta) $.
242: \item for $ \beta > 2(d-1)$, $f(\beta) = 0$.
243: % \item $\sup_{\beta \in [0,\infty)} \Big( f(\beta) \neq 0 \Big) = 2(d-1)$ %
244: \item for $\beta \in (0,2(d-1)), n \in
245: \mathbb{N}$, $f_n$ satisfies $f_n(\beta) \leq L^{1/n} n^{1/n}
246: f(\beta)$, where the constant $L$ may be chosen uniformly in $\beta \in (0,2(d-1))$.
247: \end{enumerate}
248: \end{theorem}
249: \begin{theorem}\label{thmtwo}
250: ${}$
251: \flushleft
252: \begin{enumerate}
253: \item f is log-concave on the interval $(0,2(d-1))$.
254: \item Introducing $ g: (0,2(d-1)) \to [0,\infty)$ by means of the formula
255: \begin{displaymath}
256: f(\beta) = g(\beta)\frac{ (\beta +1)^{\beta +1} } { \beta^\beta },
257: \end{displaymath}
258: we have that
259: \begin{displaymath}
260: g(\beta) \left\{ \begin{array}{ll}
261: = 1 & \textrm{on $(0,\lapha]$,}\\
262: < 1 & \textrm{on $(\lapha,2(d-1))$,}
263: \end{array} \right.
264: \end{displaymath}
265: where throughout $\lapha$ denotes the value $ 1/p_c -1 $.
266: \end{enumerate}
267: \end{theorem}
268: {\bf Remark} The assertion that $g = 1$ on $(0,\lapha]$ was originally
269: proved by Delyon \cite{DelyonT}. We include here the proof of the
270: other part of the theorem. \\
271: {\bf Proof}
272: We must show that, for $\beta \in (\lapha,2(d-1))$, $g(\beta)$ is strictly less than one.
273: Let $\beta$ lie in this interval. Let $p = 1/(1 + \beta)$. Note that $p < p_c$, and that
274: \begin{eqnarray}
275: \mathbb{P}_p(\vert C(0) \vert = n) & \geq & \mathbb{P}_p(C(0) \in \Gamma_{n,\lfloor \beta n \rfloor}) \nonumber \\
276: & = & \vert \Gamma_{n,\lfloor \beta n \rfloor} \vert \frac{{\beta}^{\lfloor \beta n \rfloor}}{(1 + \beta)^{n + \lfloor \beta n \rfloor}} \nonumber \\
277: & = & \big( f_n(\beta) \big)^n \frac{{\beta}^{\lfloor \beta n \rfloor}}{(1 + \beta)^{n + \lfloor \beta n \rfloor}}. \nonumber
278: \end{eqnarray}
279: Taking logarithms yields
280: \begin{displaymath}
281: \frac{\log \mathbb{P}_p \big( \vert C(0) \vert = n \big)}{n} \geq
282: \log f_n(\beta) + \frac{\lfloor \beta n \rfloor \log \beta}{n} -
283: \Big( 1 + \frac{\lfloor \beta n\rfloor }{n} \Big) \log (1 + \beta) ,
284: \end{displaymath}
285: from which it follows that
286: \begin{equation}\label{pop}
287: \liminf_{n \to \infty}{\frac{\log \mathbb{P}_p \big( \vert C(0) \vert = n \big)}{n}} \geq \log f(\beta) + \beta \log \beta - (1+ \beta) \log (1+ \beta) .
288: \end{equation}
289: The right-hand-side of (\ref{pop}) is equal to $\log g(\beta)$, by definition.
290: The exponential decay rate for the probability of observing a large
291: cluster in the subcritical phase was established in \cite{MR86h:82045}. Since $p < p_c$, this means the
292: left-hand-side of (\ref{pop}) is negative. This implies that $g(\beta)
293: < 1$, as required. $\Box$
294:
295: \section{Critical exponents and inequalities}\label{sectthr}
296:
297: We introduce two scaling hypotheses, each of which proposes the existence of a
298: critical exponent. We then state and prove the first
299: main theorem, which demonstrates that a pair of inequalities involving
300: the two exponents cannot both be satisfied. \\
301: {\bf Hypothesis $(\appak)$}
302: \begin{definition}
303: For each $n \in \mathbb{N}$, let $t_n \in (0,p_c)$ denote the least
304: value satisfying the condition
305: \begin{equation}\label{eteen}
306: \sum_{m}{\sdash t_n^n (1 - t_n)^m} = \sup_{p \in (0,p_c]}{\sum_{m}{\sdash p^n (1 - p)^m}}.
307: \end{equation}
308: \end{definition}
309: That is, $t_n$ is some point at or below the critical value at which the probability of observing an $n$-edged animal as the cluster to which the origin belongs is maximal. It is reasonable to suppose that $t_n$ is slightly less than $p_c$, and that the difference decays polynomially in $n$ as $n$ tends to infinity.
310:
311: \begin{definition}
312: Define $\Omega_{+}^{\appak} = \{ \beta \geq 0 : \liminf_{n \to \infty}{(p_c - t_n)/n^{-\beta}} = \infty \}$, and
313: $\Omega_{-}^{\appak} = \{ \beta \geq 0 : \limsup_{n \to \infty}{(p_c - t_n)/n^{-\beta}} = 0 \}$. \\
314: If $\sup{\Omega_{-}^{\appak}} = \inf{\Omega_{+}^{\appak}}$, then hypothesis $(\appak)$ is said to hold, and $\appak$ is defined to be equal to the common value.
315: \end{definition}
316: So, if hypothesis $(\appak)$ holds, then $p_c - t_n$ behaves like
317: $n^{-\appak}$, for large $n$. We remark that it would be consistent
318: with the notion of a scaling window about criticality that the probability of observing
319: the cluster $C(0)$ with $n$-edges achieves its maximum on the
320: subcritical interval on a short plateau whose right-hand endpoint is
321: the critical value. If this is the case, then $t_n$ should lie at the
322: left-hand endpoint of the plateau. To be confident that $p_c - t_n$ is
323: of the same order as the length of this plateau, the definition of the
324: quantities $t_n$ could be changed, so that a
325: small and fixed constant multiples the right-hand-side of
326: (\ref{eteen}). In this paper, any proof of a statement involving the
327: exponent $\appak$ is valid if it is defined in terms of this altered
328: version of the quantities $t_n$.
329:
330: {\bf Hypothesis $(\reta)$}
331:
332: This hypothesis is introduced to describe the behaviour of $f$ for
333: values of the argument just greater than $\lapha$. Theorem
334: \ref{thmtwo} asserts that the value $\lapha$ is the greatest for which $\log f(\beta) = (\beta +1)\log(\beta + 1)- \beta \log \beta$; the function $g$ was introduced to describe how $\log f$ falls away from this function as $\beta$ increases from $\lapha$. Thus, we phrase hypothesis $(\reta)$ in terms of $g$.
335: \begin{definition}
336: Define $\Omega_{-}^{\reta} = \{ \beta \geq 0 : \liminf_{\delta \to 0}{(g(\lapha + \delta)- g(\lapha))/{\delta}^{\beta}} = 0 \}$, and
337: $\Omega_{+}^{\reta} = \{ \beta \geq 0 : \limsup_{\delta \to
338: 0}{(g(\lapha + \delta) - g(\lapha))/{\delta}^{\beta}} = - \infty \}$. \\
339: If $\sup{\Omega_{-}^{\reta}} = \inf{\Omega_{+}^{\reta}}$, then hypothesis $(\reta)$ is said to hold, and $\reta$ is defined to be equal to the common value.
340: \end{definition}
341: If hypothesis $(\reta)$ holds, then greater values of $\reta$ correspond to a smoother behaviour of $f$ at $\lapha$. For example, if $\reta$ exceeds $N$ for $N \in \mathbb{N}$, then $f$ is $N$-times differentiable at $\lapha$.
342: \begin{theorem}\label{thmthr}
343: Suppose that hypotheses ($\reta$) and ($\appak$) hold. If $\appak < 1/2$, then $\reta \appak \geq 1$.
344: \end{theorem}
345: {\bf Proof}
346: We prove the Theorem by contradiction, assuming that the two
347: hypotheses hold, and that $\appak < 1/2$, $ \reta \appak < 1$. We will arrive at the conclusion that the mean cluster size, given by
348: $\sum_{n}{n \mathbb{P}_{p}{(\vert C(0) \vert = n)}}$, is bounded
349: above, uniformly for $p \in (0,p_c)$. That this is not so is proved in \cite{MR86h:82045}.
350: Note that
351: \begin{displaymath}
352: \sup_{p \in (0,p_c)}{\sum_{n}{n \mathbb{P}_{p}{(\vert C(0) \vert = n)}}} \leq \sum_{n}{n \mathbb{P}_{t_n}{(\vert C(0) \vert = n)}}.
353: \end{displaymath}
354: We write
355: \begin{equation}\label{enst}
356: \mathbb{P}_{t_n}{(\vert C(0) \vert = n)} = \sum_{m}{\sigma_{n,m} t_n^n(1 - t_n)^m},
357: \end{equation}
358: and split the sum on the right-hand-side of (\ref{enst}). To do so, we use the following definition.
359: \begin{definition}\label{alphn}
360: For $n \in \mathbb{N}$, let $\lapha_n$ be given by $t_n = 1/(1 + \lapha_n)$.
361: For $G \in \mathbb{N}$, let $D_n ( = D_n(G))$ denote the interval
362: \begin{displaymath}
363: D_n = ( \lapha_n - G{\{\log(n)/n \}}^{1/2} , \lapha_n + G{\{\log(n)/n \}}^{1/2}).
364: \end{displaymath}
365: \end{definition}
366: Now,
367: \begin{displaymath}
368: \sum_{m}{\sigma_{n,m} t_n^n(1 - t_n)^m} = C_1(n) + C_2(n) + C_3(n) ,
369: \end{displaymath}
370: where the terms on the right-hand-side are given by
371: \begin{eqnarray}
372: C_1(n) & = & \sum_{m \in n D_n}{\sigma_{n,m}t_n^n(1 - t_n)^m} , \nonumber \\
373: C_2(n) & = & \sum_{m \in n \big( (0,2(d-1)) - D_n
374: \big)}{\sigma_{n,m}t_n^n(1 - t_n)^m} \nonumber
375: \end{eqnarray}
376: and
377: \begin{displaymath}
378: C_3(n) = \sum_{m \in \{ 2(d-1)n ,\ldots, 2(d-1)n + 2d \}}{\sigma_{n,m}t_n^n(1 - t_n)^m}.
379: \end{displaymath}
380: \begin{definition}\label{defnphi}
381: Let the function $\phi: (0,\infty)^2 \to \mathbb{R}$ be given by
382: \begin{displaymath}
383: \phi (\lapha,\beta) = (\beta + 1)\log(\beta + 1) - \beta \log \beta + \beta \log \lapha - (\beta + 1)\log(\lapha + 1).
384: \end{displaymath}
385: \end{definition}
386: {\bf Remark.} That $\phi \leq 0$ is straigtforward.
387: \begin{lemma}\label{lemfour}
388: The function $\phi$ satisfies
389: \begin{displaymath}
390: \phi \big( \lapha , \lapha + \gamma \big) = - \frac{\gamma^2}{2\lapha(\lapha + 1)} + O(\gamma^3).
391: \end{displaymath}
392: \end{lemma}
393: The trivial proof is omitted. \\
394: We have that
395: \begin{eqnarray}
396: \sum_{n}{C_2(n)}
397: & = & \suf{m \in n((\lapha,2(d-1)) - D_n)}{{\bigg( f_n(m/n)\frac{\lapha_n^{m/n}}{(1 + \lapha_n)^{1 + m/n}} \bigg)}^n} \nonumber \\
398: & \leq & L \sum_{n}{ n \sum_{m \in n((\lapha,2(d-1)) - D_n)}{\exp{\{n\phi_{\lapha_n,m/n}\}}}}, \nonumber
399: \end{eqnarray}
400: where the inequality is valid by virtue of Theorem \ref{thmone} and
401: the fact that $g \leq 1$. Lemma \ref{lemfour} implies that
402: \begin{displaymath}
403: \sum_{m \in n((\lapha,2(d-1)) - D_n)}{\exp{\{n\phi_{\lapha_n,m/n}\}}} \leq (2(d-1) - \lapha) n^{-K} ,
404: \end{displaymath}
405: where $K$ may be chosen to be arbitrarily large by an appropriate choice of $G$. It is this consideration that determines the choice of $G$.
406: The miscellaneous term $C_3$ is treated by the following lemma.
407: \begin{lemma}\label{lemmanex}
408: There exists $r \in (0,1)$ such that, for $n$ sufficiently large and for $m \in \{ 2(d-1)n, \ldots, 2(d-1)n + 2d \}$, we have that
409: \begin{displaymath}
410: \sigma_{n,m} \leq \frac{ ( 1+ \frac{m}{n} )^{n+ m}}{(\frac{m}{n})^{m}} r^n .
411: \end{displaymath}
412: \end{lemma}
413: {\bf Proof} See \cite{techrep}.\\
414: We find that the $m$-indexed summand in $C_3(n)$ is at most $r^n
415: \exp{n \phi_{\lapha_n,m/n}}$: thus $C_3(n) \leq (2d+1) r^n$.
416: Note that $C_1$ satisfies
417: \begin{eqnarray}
418: C_1(n) & = & \sum_{m \in n D_n^{*}}{{\bigg( f_n(m/n)\frac{\lapha_n^{m/n}}{(1 + \lapha_n)^{1 + m/n}} \bigg)}^n} \nonumber \\
419: & \leq & L n \sum_{m \in n D_n}{g(m/n)^n \exp (n \phi_{\lapha_n,m/n})}, \nonumber
420: \end{eqnarray}
421: where the inequality is a consequence of Theorems \ref{thmone} and \ref{thmtwo}. The fact that the function $\phi$ is nowhere positive implies that
422: \begin{displaymath}
423: C_1(n) \leq L n \sum_{m \in n D_n}{g(m/n)^n}.
424: \end{displaymath}
425: Hence the desired contradiction will be reached if we can show that
426: \begin{equation}\label{expf}
427: \sum_{n}{ n \sum_{m \in n D_n}{g(m/n)^n}}
428: \end{equation}
429: is finite.
430: As such, the proof is completed by the following lemma.
431: \begin{lemma}
432: Assume hypotheses $(\reta)$ and $(\appak)$. Suppose that $\appak < 1/2$ and that $\reta \appak < 1$. Then, for $\epsilon \in (0,1 - \reta \appak)$ and $n \in \mathbb{N}$ sufficiently large,
433: \begin{equation}\label{wrte}
434: \sum_{m \in n D_n}{g(m/n)^n} \leq \exp{-n^{1 - \reta \appak - \epsilon}}.
435: \end{equation}
436: \end{lemma}
437: {\bf Proof}
438: Let ${\reta}^* > \reta$ and ${\appak}^* > \appak$ be such that
439: $\appak^* < 1/2$ and ${\reta}^* {\appak}^* < \reta \appak + \epsilon$.
440: By hypothesis $(\reta)$, there exists $\epsilon'>0$ such that
441: \begin{displaymath}
442: \delta \in (0,{\epsilon}') \ \textrm{implies} \ g(\lapha + \delta) - g(\lapha) < - {\delta}^{{\reta}^*}.
443: \end{displaymath}
444: From Theorems \ref{thmone} and \ref{thmtwo}, it follows that $\sup_{\beta \in [\lapha +
445: \epsilon',2(d-1)]}{g(\beta)} < 1$, which shows that the contribution
446: to the sum in (\ref{wrte}) from all those terms indexed by $m$ for
447: which $m/n > \lapha + \epsilon'$ is exponentially decaying in
448: $n$. Thus, we may assume that
449: there exists $N_1$ such that for $n \geq N_1$, if $m \in D_n^{*}$ then $m/n - \lapha < \epsilon'$. Note that, by hypothesis $(\appak)$, $\lapha_n - \lapha \geq n^{-{\appak}^*}$ for sufficiently large. Hence, there exists $N_2$ such that, for $n \geq N_2$,
450: \begin{displaymath}
451: \lapha_n - G(\log(n)/n)^{1/2} \geq \lapha + n^{-{\appak}^*} - G(\log(n)/n)^{1/2} \geq \lapha + (1/2)n^{-{\appak}^*}.
452: \end{displaymath}
453: For $n \geq \max\{N_1,N_2\}$ and $m \in n D_n^*$,
454: \begin{eqnarray}
455: g(m/n) & \leq & 1 - (m/n - \lapha)^{{\reta}^*} \nonumber \\
456: & \leq & 1 - (\lapha_n - G(\log(n)/n)^{1/2} - \lapha)^{{\reta}^*} \nonumber \\
457: & \leq & 1 - ((1/2)n^{-{\appak}^*})^{{\reta}^*}. \nonumber
458: \end{eqnarray}
459: So, for $n \geq max(N_1,N_2)$,
460: \begin{displaymath}
461: \sum_{m \in n D_n^{*}}{g(m/n)^n} \leq (2G(n \log(n))^{1/2})[1 - C' n^{-{\appak}^* {\reta}^*}]^{n},
462: \end{displaymath}
463: for some constant $C' > 0$.
464: There exists $g \in (0,1)$, such that for large $n$,
465: \begin{displaymath}
466: [1 - C' n^{-{\appak}^* {\reta}^*}]^{n} \leq g^{n^{1 - {\appak}^*{\reta}^*}}.
467: \end{displaymath}
468: This implies that
469: \begin{displaymath}
470: \sum_{m \in n D_n^{*}}{g(m/n)^n} \leq h^{n^{1 - {\appak}^*{\reta}^*}} \ \textrm{for large $n$ and $h \in (g,1)$}.
471: \end{displaymath}
472: From ${\reta}^* {\appak}^* < \reta \appak + \epsilon$, we find that
473: \begin{displaymath}
474: \sum_{m \in n D_n^{*}}{g(m/n)^n} \leq \exp{-n^{1 - \reta \appak - \epsilon}} \ \textrm{for large $n$,}
475: \end{displaymath}
476: as required. $\Box$
477: \section{Sufficient conditions for $\theta(p_c)=0$}\label{sectfour}
478: In this section, we give two theorems, demonstrating sufficient
479: conditions for the continuity of the percolation probability in terms
480: of inequalities on $\reta$ and $\appak$.
481: \begin{theorem}\label{thmfour}
482: Assume that hypotheses ($\reta$) and ($\appak$) hold.
483: \flushleft
484: \begin{enumerate}
485: \item Suppose that $\reta < 2$. Then $\theta(p_c) = 0$.
486: \item Suppose that $\appak > 1/2$. Then $\theta(p_c) = 0$.
487: \end{enumerate}
488: \end{theorem}
489: The proof of Theorem \ref{thmfour} will exploit the characterisation of continuity provided by the following lemma.
490: \begin{definition}
491: ${}$
492: \flushleft
493: \begin{itemize}
494: \item Let $\sigma(p)= \suf{m}{\sigm}$.
495: \item Let $\sigma_N(p)=\sum_{n \leq N}{\sum_{m}{\sigm}}$
496: \end{itemize}
497: \end{definition}
498: \begin{lemma}\label{lemghe}
499: A necessary and sufficient condition for $\theta(p_c)=0$ is that $\sigma_n$ tends uniformly to $\sigma$ on the interval $(0,p_c)$.
500: \end{lemma}
501: {\bf Proof} See \cite{techrep}. \\
502: {\bf Proof of Theorem \ref{thmfour}}
503: By Lemma \ref{lemghe}, to establish that $\theta(p_c)=0$, it suffices
504: to show that $\sigma_n$ tends to $\sigma$ uniformly on $(0,p_c)$. We
505: begin by verifying this condition under the hypotheses of the first
506: part of the Theorem. We will show that
507: \begin{equation}\label{escc}
508: \sum_{n}{\sum_{m}{\sigma_{n,m}\sup_{p \in (0,p_c)}p^n(1-p)^m}} < \infty .
509: \end{equation}
510: This will do because
511: \begin{eqnarray}
512: \sup_{p \in (0,p_c)}{\big( \sigma(p) - \sigma_{N}(p) \big)} & = & \sup_{p \in (0,p_c)}{\sum_{n \geq N+1}{\sum_{m}\sigm}} {}\nonumber\\
513: & \leq & \sum_{n \geq N+1}{\sum_{m}{\sdash \sup_{p \in (0,p_c)}{p^n(1-p)^m}}} {}\nonumber
514: \end{eqnarray}
515: So the condition stated in (\ref{escc}) implies the uniform convergence of $\sigma_n$ to $\sigma$ on the subcritical interval.
516:
517: Note that
518: \begin{displaymath}
519: \sup_{p \in (0,p_c)}{p^n(1-p)^m} = \left\{ \begin{array}{ll} \sigt & \textrm{for $n/(n+m) \leq p_c$} \\
520: \sigo & \textrm{for other pairs $(n,m)$}. \end{array} \right.
521: \end{displaymath}
522:
523: This observation allows us to decompose the sum appearing in (\ref{escc}):
524:
525: \begin{eqnarray}
526: \suf{m}{\sdash \sup_{p \in (0,p_c)}{p^n(1-p)^m}} & = & \suh{m =
527: 1}{\lfloor n \lapha \rfloor }{\sigma_{n,m} \sigo } \\ \label{thisa}
528: & + & \suf{m > \lfloor n \lapha \rfloor}{\sigma_{n,m}\sigt} .{}\nonumber
529: \end{eqnarray}
530: Now,
531: \begin{displaymath}
532: \suh{m = 1}{\lfloor n \lapha \rfloor}{\sdash \sigo } \leq \suf{m}{\sdash \sigo},
533: \end{displaymath}
534: which is less than or equal to one, being the critical probability that the origin lies in a finite cluster.
535:
536: Set $A$ equal to the second sum on the right-hand-side of (\ref{thisa}).
537: It suffices to show that $A$ is finite. Our strategy is to split each of the summands of $n$ into two parts, each of which is a sum over $m$ in an interval which has an $n$-dependence. The first sum, $A_1$, will include those $m$-values sufficiently close to $n \lapha$ that this term can be bounded in terms of the critical probability of observing a large cluster. The second sum, $A_2$, will be shown to decay quickly, under the assumption that $\reta < 2$.
538:
539: Write $A = A_1 + A_2$, where
540: \begin{eqnarray}
541: A_1 & = & \suh{m = \lfloor n \lapha \rfloor + 1}{\lfloor n \lapha + n^{1/2} \rfloor +1 }{\sdash\sigt}, \nonumber \\
542: \textrm{and} \, \, A_2 & = & \suf{m > \lfloor n \lapha + n^{1/2} \rfloor +1}{\sdash\sigt}.\nonumber
543: \end{eqnarray}
544: Recalling that $\lapha = 1/p_c - 1$,
545: \begin{displaymath}
546: A_1 = \suh{m = \lfloor n \lapha \rfloor + 1}{\lfloor n
547: \lapha + n^{1/2} \rfloor + 1 }{\sdash\sigo \exp\big( - n \phi (\lapha,m/n) \big)},
548: \end{displaymath}
549: where the function $\phi$ was specified in Definition \ref{defnphi}.
550: For each $m \in \{ \lfloor n \lapha \rfloor, \ldots, \lfloor n \lapha
551: + n^{1/2} \rfloor + 1 \}$, $c_m \in (0,3/2)$, where $c_m$ is given by $m/n = \lapha + c_m n^{-1/2}$.
552: Lemma \ref{lemfour} implies that for any sufficiently large $C'$, there exists $N_1$ such that for all $n \geq N_1$, and for $m \in \{ \lfloor n \lapha \rfloor + 1, \ldots, \lfloor n \lapha + n^{1/2} \rfloor + 1 \}$,
553: \begin{displaymath}
554: - \phi (\lapha,m/n) \leq 9/[8 n \lapha(\lapha +1)] + C'/{n^{3/2}} .
555: \end{displaymath}
556: From this, we deduce that for $n \geq N_1$ and $m \in \{
557: \lfloor n \lapha \rfloor + 1, \ldots, \lfloor n \lapha + n^{1/2}
558: \rfloor + 1 \}$, \mbox{$\exp{(- n \phi (\lapha,m/n) )}$} is bounded above, by $C$, say.
559: So,
560: \begin{eqnarray}
561: A_1 & \leq & \sum_{n < N_1}{\sum_{m \in \{ \lfloor n \lapha \rfloor + 1, \ldots, \lfloor n \lapha + n^{1/2} \rfloor + 1 \}}{\sdash\sigt}} {} \nonumber \\
562: {} & & + C \sum_{n \geq N_1}{\sum_{m \in \{ \lfloor n \lapha \rfloor + 1, \ldots, \lfloor n \lapha + n^{1/2} \rfloor + 1 \}}{\sdash\sigo}}, \nonumber
563: \end{eqnarray}
564: which is finite, as desired.
565:
566: We now seek to bound $A_2$:
567: \begin{eqnarray}
568: A_2 & = & \suf{\atwo}{{\bigg(f_n(m/n)\Big(\frac{n}{n+m}\Big){\Big(\frac{m}{n+m}\Big)}^{m/n}\bigg)}^n} {} \nonumber \\
569: & \leq & L \sum_{n}{\sum_{m = \lfloor n \lapha + n^{1/2} \rfloor +
570: 2}^{2(d-1)n - 1}{ n
571: \bigg(f(m/n) \Big(\frac{n}{n+m}\Big){\Big(\frac{m}{n+m}\Big)}^{m/n}\bigg)^n }} \nonumber \\
572: & & + {} \sum_{n}{\sum_{m = 2(d-1)n}^{2(d-1)n + 2d}{\sdash\sigt}} \nonumber
573: \end{eqnarray}
574: where the inequality follows from Theorem \ref{thmone} and the fact
575: that $g \leq 1$. By Lemma \ref{lemmanex}, there exists $r \in (0,1)$ such that, for $n$ sufficiently large,
576: \begin{displaymath}
577: \sum_{m = 2(d-1)n}^{2(d-1)n + 2d}{\sdash\sigt} \leq (2d+1) r^n .
578: \end{displaymath}
579: It follows from the definition of the function $g$ that
580: \begin{equation}\label{eat}
581: A_2 \leq L \sum_{n}{\sum_{m = \lfloor n \lapha + n^{1/2} \rfloor + 2}^{2(d-1)n - 1}{n g(m/n)^n}} + (2d+1) \sum_{n}{r^n}.
582: \end{equation}
583: To bound the first term in the expression on the
584: right-hand-side of (\ref{eat}),
585: let $\epsilon \in (0,2 - \reta)$. Let ${\delta}' > 0$, be such that, for $\delta \in (0,{\delta}')$, $g(\lapha + \delta) - g(\lapha) < -{\delta}^{\reta + \epsilon}$. Let $\gamma \in (0,1)$ be such that
586: \begin{displaymath}
587: \sup_{\beta \in (\lapha + {\delta }',2(d-1))}{g(\beta)} < \gamma .
588: \end{displaymath}
589: Note that
590: \begin{eqnarray}
591: & & \suh{m = \lfloor n \lapha + n^{1/2} \rfloor
592: +2}{\lfloor n(\lapha + {\delta}') \rfloor}{ n g(m/n)^n} {} \nonumber \\
593: & \leq & \suh{m = \lfloor n \lapha + n^{1/2} \rfloor
594: +2}{\lfloor n(\lapha + {\delta}') \rfloor}{ n \Big( 1 - ( m/n -
595: \lapha)^{\reta + \epsilon} \Big)^n} {} \nonumber \\
596: & \leq & {\delta}' \sum_{n}{n^2 \big( 1 - n^{ - \frac{\reta +
597: \epsilon}{2}} \big)^n}. \nonumber
598: \end{eqnarray}
599: Since $\reta + \epsilon < 2$, this expression is finite.
600: Note also that
601: \begin{displaymath}
602: \suh{m = \lfloor n(\lapha + {\delta}') \rfloor +
603: 1}{2(d-1)n - 1}{n g(m/n)^n}
604: \leq 2(d-1)\sum_{n}{n^2 \gamma^n} < \infty .
605: \end{displaymath}
606: We deduce that $A_2$ is finite and in doing so, complete the proof of
607: the first part of Theorem \ref{thmfour}.
608:
609: We now prove the second part of the Theorem.
610: A sufficient condition for continuity is
611: \begin{equation}\label{anothscc}
612: \suf{m}{\sdash t_n^n (1 - t_n)^m} < \infty .
613: \end{equation}
614: Indeed, the supremum over $p$ in $(0,p_c)$ of $\sigma - \sigma_N$ is
615: bounded above by the expression in (\ref{anothscc}) with the sum in $n$
616: being taken over values exceeding $N - 1$. By Lemma \ref{lemghe}, if
617: (\ref{anothscc}) holds, then $\theta(p_c)=0$.
618:
619: The fact that $t_n \leq p_c$ implies that $t_n^n(1 - t_n)^m \leq
620: p_c^n(1 - p_c)^m$ provided that $n/(n+m) > p_c$, which holds if and
621: only if $m \leq \lfloor n \lapha \rfloor$. From this, we may deduce that
622: \begin{eqnarray}
623: \suh{m = 1}{\lfloor n \lapha \rfloor}{\sdash t_n^n(1 - t_n)^m}
624: & \leq & \suh{m = 1}{\lfloor n \lapha \rfloor}{\sdash \sigo} \nonumber \\
625: & \leq & \suf{m}{\sdash \sigo} \leq 1 \nonumber
626: \end{eqnarray}
627: To verify the condition in (\ref{anothscc}), we must bound the expression
628: \begin{equation}\label{wret}
629: \suh{m = \lfloor n \lapha \rfloor + 1}{2(d-1)n + 2d}{\sdash t_n^n(1 - t_n)^m}.
630: \end{equation}
631: To do so, we make the following definition.
632: \begin{definition}
633: For $G \in \mathbb{N}$, let $D_n^{*} ( = D_n^{*}(G))$ denote the interval
634: \begin{displaymath}
635: D_n^{*} = (\max{\{ \lapha, \lapha_n - G{\{\log(n)/n \}}^{1/2} \}}, \lapha_n + G{\{\log(n)/n \}}^{1/2}),
636: \end{displaymath}
637: where the constants $\{ \lapha_n : n \in \mathbb{N} \}$ were specified
638: in Definition \ref{alphn}.
639: \end{definition}
640: Allowing that $G$ will be determined slightly later, we write the
641: expression in (\ref{wret}) in the form
642: \begin{eqnarray}
643: & & \suf{m \in n D_n^{*}}{\sdash t_n^n(1 - t_n)^m} + \suf{m \in n((\lapha,2(d-1)) - D_n^{*})}{\sdash t_n^n(1 - t_n)^m} \nonumber \\
644: & & \, + \, \sum_{n}{\sum_{m = 2(d-1)n}^{2(d-1)n + 2d}{\sdash t_n^n(1 - t_n)^m}} . \label{fvg}
645: \end{eqnarray}
646: An argument identical to that by which the term $C_2$ was bounded in the
647: proof of Theorem \ref{thmthr} yields
648: \begin{displaymath}
649: \suf{m \in n((\lapha,2(d-1)) - D_n^{*})}{\sdash t_n^n(1 - t_n)^m}
650: \leq \sum_{n}{n^{-K}},
651: \end{displaymath}
652: where $K$ may be chosen to be arbitrarily large by an appropriate
653: choice of $G$, thereby determining how $G$ is chosen. The third term
654: in (\ref{fvg}) was labelled $C_3(n)$ in the proof of Theorem \ref{thmthr}
655: and was shown to be bounded above by $(2d+1) r^n$ for $n$ sufficiently
656: high.
657: We have that
658: \begin{eqnarray}
659: & & \suf{m \in n D_n^{*}}{\sdash t_n^n(1 - t_n)^m} \nonumber \\
660: & = & \suf{m \in n D_n^{*}}{\sdash \frac{\lapha^m}{(1 + \lapha)^{n+m}}
661: \exp{n\Phi(\lapha_n,\lapha,m/n)} }, \label{kzc}
662: \end{eqnarray}
663: where
664: \begin{eqnarray}
665: \Phi(\gamma,\lapha,\beta) & = & \beta \log \gamma - (\beta + 1) \log ( \gamma + 1) - \beta \log \lapha + (\beta + 1)\log(\lapha + 1) \nonumber \\
666: & = & \beta \log (1 + (\gamma - \lapha)/\lapha ) - (\beta + 1) \log (1 + (\gamma - \lapha)/(1+ \lapha)) \nonumber \\
667: & = & - \, \, \frac{(\gamma - \lapha)^2}{2} [\beta/{\lapha}^2 - (\beta + 1)/(1 + \lapha)^2] \nonumber \\
668: & & + \, \, \frac{(\gamma - \lapha)(\beta - \lapha)}{\lapha(\lapha + 1)} \, + \, O\big[(\gamma - \lapha)^3 \big]. \nonumber
669: \end{eqnarray}
670: We are supposing that hypothesis $(\appak)$ holds, and that $\appak > 1/2$. Let ${\appak}'$ satisfy $\appak > {\appak}' > 1/2$.
671: In this context,
672: \begin{displaymath}
673: \Phi(\lapha_n,\lapha,\beta) = \frac{(\lapha_n - \lapha)(\beta - \lapha)}{\lapha(\lapha + 1)} - \frac{(\lapha_n - \lapha)^2}{2} [\beta/{\lapha}^2 - (\beta + 1)/(1 + \lapha)^2] + O(n^{-3{\appak}'}).
674: \end{displaymath}
675: Now, $\beta \in D_n^{*}$ implies that there exists $C' > 0$ such that
676: $\beta - \lapha \leq C'n^{-{\appak}'} +C'(\log(n)/n)^{1/2}$ ; since
677: $\appak' > 1/2$, we may write $\beta - \lapha \leq
678: C'(\log(n)/n)^{1/2}$, where the value of $C'$ has been increased if
679: necessary. For such $\beta$, $\Phi(\lapha_n,\lapha,\beta) \leq
680: C'n^{-{\appak}' - {1/2}}{\log(n)}^{1/2} + n^{-2{\appak}'} +
681: O(n^{-3{\appak}'})$. This implies that, for all $n$ and $\beta \in
682: D_n^{*}$, $\exp{ n \Phi(\lapha_n,\lapha,\beta)} < C'$, where once again the value of $C'$ may have changed.
683: Recalling that $\lapha = 1/p_c - 1$, we deduce from (\ref{kzc}) that
684: \begin{eqnarray}
685: & & \suf{m \in n D_n^{*}}{\sdash t_n^n(1 - t_n)^m} \nonumber \\
686: & \leq & C' \suf{m \in n D_n^{*}}{\sdash \sigo} \nonumber \\
687: & \leq & C' \suf{m}{\sdash \sigo} \leq C', \nonumber
688: \end{eqnarray}
689: proving the second part of Theorem \ref{thmthr}.$\Box$ \\
690: We now examine the case where
691: $\appak < 1/2$ and $\reta \appak > 1$.
692: \begin{definition}
693: Let $n \in \mathbb{N}$, and $\beta \in (0,2(d-1))$. Set
694: \begin{equation}\label{eanb}
695: a_n(\beta) = {\bigg( \frac{f_n'(\beta)}{f(\beta)} \bigg) }^n.
696: \end{equation}
697: \end{definition}
698: {\bf Remark} The quantities $a_n(\beta)$ appear in the factorisation of $\sigma_{n,\lfloor \beta n \rfloor}$,
699: \begin{displaymath}
700: \sigma_{n,\lfloor \beta n \rfloor} = a_n(\beta)g(\beta)^n \bigg( \frac{(\beta + 1)^{\beta + 1}}{\beta^\beta} \bigg)^n.
701: \end{displaymath}
702: As such, they measure the extent to which the exponential growth rate
703: $f(\beta)$ is underestimated by $\sigma_{n,m}$.
704:
705: Performing a similar analysis to that undertaken during each part of
706: Theorem \ref{thmfour} yields the following result. Its proof appears in \cite{techrep}.
707: \begin{theorem}\label{thmund}
708: Assume that hypotheses $(\reta)$ and $(\appak)$ hold. Suppose that \mbox{$\appak < 1/2$} and \mbox{$\reta \appak > 1$}. Let $K$ be large. Then there exist constants $\epsilon > 0$ and $C > 0$ such that for each $n \in \mathbb{N}$,
709: \begin{eqnarray}
710: \epsilon \sum_{\alplow}{a_n(m/n)} & \leq & \sum_{m}{\sdash \sigo} \label{edf} \\
711: & \leq & \sum_{\alphigh}{a_n(m/n)}
712: + \ n^{-K} \nonumber
713: \end{eqnarray}
714: and
715: \begin{eqnarray}
716: \epsilon \sum_{\alpnlow}{a_n(m/n)} & \leq & \sum_{m}{\signt} \label{edg} \\
717: & \leq &
718: \sum_{\alpnhigh}{a_n(m/n)} \ + \ n^{-K} . \nonumber
719: \end{eqnarray}
720: \end{theorem}
721: {\bf Remark} Here, $B(a,b)$ denotes the interval $(a-b,a+b)$. Note
722: also that it follows from Theorem \ref{thmund} that the
723: condition
724: \begin{displaymath}
725: \sum_{\alpnhigh}{a_n(m/n)} < \infty
726: \end{displaymath}
727: implies that $\theta(p_c)=0$, without recourse to scaling hypotheses. In examining this condition, bounds on the entropic exponent are revelant (see \cite{MR95m:82076}).
728:
729: \section{Scaling law}\label{sectfive}
730: In this section, we examine the exponential decay rate in $n$ for the
731: probability of the
732: event $\{ C(0)=n \}$ for $p$ slightly less than $p_c$ by our combinatorial
733: approach. In doing so, we relate the quantity $\reta$ to the exponent
734: for correlation size, and see how the scaling behaviour for the typical
735: surface-area-to-volume ratio of unusually large clusters in the
736: marginally subcritical regime depends on the value of $\reta$.
737: \begin{definition}
738: Let $q:(0,p_c) \to [0,\infty)$ be given by
739: \begin{displaymath}
740: q(p) = \lim_{n \to \infty}{\frac{- \log \mathbb{P}_p(\vert C(0) \vert = n)}{n}}.
741: \end{displaymath}
742: Define
743: $\Omega_{+}^{\rrho} = \{ \gamma \geq 0 : \liminf_{p \uparrow p_c} \frac{q(p)}{(p_c - p)^{\gamma}} = \infty \}$ \ \textrm{and}\\
744: $\Omega_{-}^{\rrho} = \{ \gamma \geq 0 : \limsup_{p \uparrow p_c}{ \frac{q(p)}{(p_c - p)^{\gamma}}} = 0 \}$.
745: If $\sup{\Omega_{-}^{\rrho}} = \inf{\Omega_{+}^{\rrho}}$, then hypothesis $(\rrho)$ is said to hold, and $\rrho$ is defined to be equal to the common value.
746: \end{definition}
747: {\bf Remark} The existence of $q$ follows from a standard
748: subadditivity argument. \\
749: The quantity $\rrho$ might reasonably be called the exponent for
750: `correlation size'.
751: \begin{theorem}\label{thmfive}
752: There exists $\delta' > 0$ and $p_0 \in (0,p_c)$ such that $p \in
753: (p_0,p_c)$ implies that $q(p)$ is given by
754: \begin{displaymath}
755: \inf_{\beta \in (\lapha,\lapha + \delta')}{- \log g(\beta) - \phi
756: \big( 1/p - 1,\beta \big)}.
757: \end{displaymath}
758: \end{theorem}
759: The proof, whose details are given in \cite{techrep}, relies on the
760: fact that the probability that the cluster $C(0)$ has $n$ edges and
761: $m$ outlying edges in a percolation with parameter $p$ is given by
762: $$
763: a_n(m/n) \exp{n \Big( \log g(m/n) + \phi
764: \big( 1/p - 1,\beta \big) \Big)},
765: $$
766: the first term $a_n(m/n)$ having subexponential decay for large
767: $n$.
768:
769: Theorem \ref{thmfive} allows us to deduce a scaling law that relates the combinatorially defined exponent $\reta$ to one which is defined directly from the percolation model.
770: \begin{theorem}\label{thmsix}
771: Assume hypothesis $(\reta)$.
772: \begin{itemize}
773: \item Suppose that $\reta \in (1,2)$. Then hypothesis $(\rrho)$ holds and
774: $\rrho = 2$.
775: \item Suppose that $\reta \in (2,\infty)$. Then hypothesis $(\rrho)$
776: holds and $\rrho = \reta$.
777: \end{itemize}
778: \end{theorem}
779: {\bf Proof}
780: Suppose that $\reta \in (1,2)$. Choose $\epsilon > 0$ so that $1 < \reta
781: - \epsilon < \reta + \epsilon < 2$. There exists constants $C_1,C_2>0$
782: such that, for $p \in (p_0,p_c)$ and $\beta \in (\lapha,\lapha +
783: \delta')$,
784: \begin{eqnarray}
785: (\beta - \lapha)^{\reta + \epsilon} + C_1 \big( \beta - (1/p - 1)
786: \big)^2 & \leq & - \log g(\beta) + - \phi \big( 1/p - 1,\beta \big) \label{rtf} \\
787: & \leq & (\beta - \lapha)^{\reta - \epsilon} + C_2 \big( \beta - (1/p - 1)
788: \big)^2. \nonumber
789: \end{eqnarray}
790: Applying Theorem \ref{thmfive}, we find that
791: \begin{equation}\label{rtfo}
792: (\beta_p - \lapha)^{\reta + \epsilon} + C_1 \big( \beta_p - (1/p - 1)
793: \big)^2 \leq q(p),
794: \end{equation}
795: where $\beta_p \in [\lapha,\lapha +
796: \delta']$ denotes a value at which the infimum in the interval $[\lapha,\lapha +
797: \delta']$ of the first term in (\ref{rtf}) is
798: attained.
799: Let $y_p = 1/p - 1 - \lapha$, and let $\sigma_p$ satisfy $\beta_p = \lapha +
800: y_p^{\sigma_p}$. Then $\beta_p$ and $\sigma_p$ satisfy
801: \begin{eqnarray}
802: (\reta + \epsilon) (\beta_p - \lapha)^{\reta + \epsilon - 1} & = & - 2 C_1
803: \big( \beta_p - (1/p - 1) \big) \nonumber \\
804: (\reta + \epsilon) y_p^{\sigma_p ( \reta + \epsilon - 1)} & = & 2 C_1
805: \big( y_p - y_p^{\sigma_p} \big) \label{ghb}
806: \end{eqnarray}
807: Since $\beta_p \leq 1/p - 1$, $\sigma_p \geq 1$. From this and
808: (\ref{ghb}) follows $\liminf_{p \uparrow p_c}{\sigma_p} \geq
809: 1/(\reta + \epsilon - 1)$. Applying (\ref{ghb}) again, we deduce that
810: $\lim_{p \uparrow p_c}{\sigma_p} =
811: 1/(\reta + \epsilon - 1)$.
812: Substituting $\sigma_p$ in (\ref{rtfo}) yields
813: \begin{displaymath}
814: y_p^{\sigma_p (\reta + \epsilon)} + C_1 \big( y_p - y_p^{\sigma_p}
815: \big)^2 \leq q(p).
816: \end{displaymath}
817: The facts that $\lim_{p \uparrow}{\sigma_p} > 1$ and $\lim_{p
818: \uparrow}{\sigma_p (\reta + \epsilon)} = (\reta + \epsilon)/(\reta +
819: \epsilon - 1) > 2$ imply that, for a small constant $c$, \mbox{$c (p_c
820: - p)^2 \leq q(p)$} for values of $p$ just less than $p_c$. A similar
821: analysis in which $q(p)$ is bounded below by the infimum on the
822: interval $[\lapha,\lapha + \delta']$ of the third expression in
823: (\ref{rtf}) implies that for large $C$, \mbox{$q(p) \leq C (p_c
824: - p)^2$}, in a similar range of values of $p$. Thus hypothesis $(\rrho)$ holds, and $\rrho = 2$.
825:
826: In the case where $\reta > 2$, let $\epsilon > 0$ be such that $\reta >
827: 2 + \epsilon$. Defining $\sigma'_p$ by \mbox{$\beta_p = 1/p - 1 -
828: y_p^{\sigma'_p}$}, we find that
829: \begin{equation}\label{qew}
830: (\reta + \epsilon)
831: \big( y_p - y_p^{\sigma'_p} \big)^{\reta + \epsilon - 1} = 2 C_1
832: y_p^{\sigma'_p}.
833: \end{equation}
834: Note that $\beta_p \geq \lapha$ implies that $\sigma'_p \geq 1$. From
835: (\ref{qew}), it follows that $\liminf_{p \uparrow
836: p_c}{\sigma'_p} \geq \reta + \epsilon - 1$. Since $\reta + \epsilon -
837: 1 > 1$, applying (\ref{qew}) again shows that the limit $\lim_{p \uparrow
838: p_c}{\sigma'_p}$ exists and infact equals $\reta + \epsilon - 1$. Substituting $\sigma'_p$ in (\ref{rtf}) yields
839: \begin{displaymath}
840: \big( y_p - y_p^{\sigma'_p}
841: \big)^{\reta+ \epsilon} + C_1 y_p^{2 \sigma'_p} \leq q(p).
842: \end{displaymath}
843: The fact that $\liminf_{p \uparrow p_c}{\sigma'_p} > 1$ implies that
844: \mbox{$c (p_c - p)^{\reta + \epsilon} \leq q(p)$} for values of $p$
845: just less than $p_c$. Making use of the
846: inequality $\reta > 2 + \epsilon$ in considering the infimum
847: of the third term appearing in (\ref{rtf}) yields in this case
848: \mbox{$q(p) \leq C (p_c - p)^{\reta - \epsilon}$} for similar values of
849: $p$. Thus, since $\epsilon$
850: may be chosen to be arbitrarily small, we find that, if $\reta > 2$, then hypothesis $(\rrho)$ holds, and
851: that $\rrho = \reta$. $\Box$ \\
852: {\bf Acknowledgements} Financial support was provided by a Domus Graduate Scholarship (Competition B) of Merton College, Oxford.
853: I would like to thank Terry Lyons
854: for stimulating and helpful discussions. I thank
855: John Cardy, Amir Dembo and Mathew Penrose for their helpful comments.
856: \bibliography{pbiblio}
857: \bibliographystyle{plain}
858: \end{document}
859:
860:
861:
862:
863:
864:
865: