1: \documentclass[12pt]{article}

2: \title{Brownian beads}

3: \author{B\'alint Vir\'ag\footnote{Department of Mathematics, MIT, Cambridge, MA

4: 02139, USA. {\sf balint@math.mit.edu}. Research partially

5: supported by NSF grant \#DMS-0206781.}}

6: \date{May 27, 2003}

7: \def\myabstract{}

8: \oddsidemargin 0in \topmargin 0in \headheight 0in \headsep 0in

9: \textheight 9in \textwidth 6.5in

10: \renewcommand{\baselinestretch}{1.3}

11: 

12: \usepackage{amsfonts}

13: \usepackage{graphicx}

14: \usepackage{amsmath}

15: \usepackage{amsthm}

16: 

17: \newtheorem{theorem}{Theorem}

18: \newtheorem{definition}[theorem]{Definition}

19: \newtheorem{question}[theorem]{Question}

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

21: \newtheorem{fact}[theorem]{Fact}

22: \newtheorem{remark}[theorem]{Remark}

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

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

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

26: \newenvironment{theorem*}[1]{\par \trivlist

27:  \itemindent 0pt \item[\hskip\labelsep\bf Theorem #1]

28:  \it\ignorespaces}{\endtrivlist}

29: \newenvironment{eg}{\par \trivlist

30:  \itemindent 0pt \item[\hskip\labelsep\bf Example

31:  \refstepcounter{theorem} \thetheorem]\ignorespaces}{\endtrivlist}

32: %\newcommand{\qed}{\hfill\mbox{$\framebox(5,5)[]{}$}}

33: \renewenvironment{proof}{\par \trivlist

34:  \itemindent\parindent \item[\hskip\labelsep\sc Proof.]

35:  \ignorespaces}{\qed\endtrivlist}

36: \newenvironment{proofof}[1]{\par \trivlist

37:  \itemindent\parindent \item[\hskip\labelsep\sc Proof of #1.]

38:  \ignorespaces}{\qed\endtrivlist}

39: \renewcommand{\theequation}{%\thesection.

40: \arabic{equation}}

41: 

42: \newcommand\mnote[1]{}

43: %\newcommand\mnote[1]{\marginpar{\ \\ \small \tt #1}}

44: \newcommand\lb[1]{\label{#1}\mnote{#1}}

45: \newcommand\bel[1]{{\mnote{#1}}\begin{equation}\label{#1}}

46: %\newcommand\bel[1]{\begin{equation}\lb{#1}}

47: \newcommand\ee{\end{equation}}

48: 

49: \newcommand\re[1]{(\ref{#1})}

50: \newcommand{\Strip}{\mathbb S}

51: \newcommand{\eps}{\varepsilon}

52: \newcommand{\Var}{\mbox{Var}}

53: \newcommand{\Z}{\mathbb Z}

54: \newcommand{\R}{\mathbb R}

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

56: \newcommand{\nis}{\hbox{$\cap$\kern-.6em\lower0em\hbox{$/$}}\,}

57: \newcommand{\C}{\mathbb C}

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

59: \newcommand{\nats}{\mathbb N}

60: \newcommand{\Zd}{{\mathbb Z}^d}

61: \newcommand{\Rd}{{\mathbb R}^d}

62: \newcommand{\HH}{{\mathbb H}}

63: \newcommand{\hull}{\mathcal H}

64: \newcommand{\UU}{\mathbb U}

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

66: \newcommand{\FF}{\mathcal F}

67: \newcommand{\GG}{\mathcal G}

68: \newcommand{\Ubar}{\overline {\mathbb U}}

69: \newcommand{\Hbar}{\overline {\mathbb H}}

70: \newcommand{\ev}{{\mathbf E}}

71: \newcommand{\pr}{{\mathbf P}}

72: \newcommand{\one}{{\mathbf 1}}

73: \newcommand{\gi}{\,|\,}

74: \newcommand{\as}{\mbox{\hspace{.3cm} a.s.}}

75: \newcommand{\dist}{\mbox{dist}}

76: \newcommand{\Ghat}{{\hat G}}

77: \newcommand{\CBM}{\sf CBM}

78: \newcommand{\BE}{\mathsf{BE}}

79: \newcommand{\bm}{\mathsf{BM}}

80: \newcommand{\BB}{\mathsf{BB}}

81: \newcommand{\nbb}{\mathsf{NBB}}

82: \newcommand{\ic}{IC}

83: \newcommand{\Rnn}{\R_{\ge 0}}

84: \newcommand{\Dbar}{{\overline D}}

85: \newcommand{\cd}{\Rightarrow}

86: \newcommand{\omu}{{\overline \mu}}

87: \newcommand{\onu}{{\overline \nu}}

88: \newcommand{\diam}{\operatorname{diam}}

89: \newcommand{\capz}{\mbox{\rm cap}_0}

90: \newcommand{\capo}{\mbox{\rm cap}_1}

91: 

92: \usepackage{natbib}

93: \bibliographystyle{balint}

94: 

95: \begin{document}

96: \maketitle

97: %\footnotetext[2]

98: %{{\it AMS} 1991 {\it subject classifications. } Primary ;

99: %secondary .}

100: %\footnotetext[3]{

101: %{\it Key words and phrases.} Brownian motion}

102: \begin{abstract} \noindent

103: We show that the past and future of half-plane Brownian motion

104: at certain cutpoints are independent of each other after a

105: conformal transformation. Like in It\^o's excursion theory, the

106: pieces between cutpoints form a Poisson process with respect to

107: a local time. The size of the path as a function of this local

108: time is a stable subordinator whose index is given by the

109: exponent of the probability that a stretch of the path has no

110: cutpoint. The index is computed and equals $1/2$.

111: \end{abstract}

112: 

113: \section{Introduction}

114: 

115: The mathematical theory of conformally invariant planar stochastic

116: models has seen great progress in the recent years. The goal of

117: this paper is to consider Brownian motion, the first example of a

118: conformally invariant process, and further explore its conformal

119: structure.

120: 

121: Let $B$ denote Brownian motion started at zero and conditioned to

122: stay in the upper half plane; we call this distribution $\BE$, or

123: half-plane excursion. This is a transient process, and he path has

124: cut-points \citep{burdzy89}, that is points which, if removed,

125: make the image of $B$ disconnected. We call the segments of the

126: paths between consecutive cutpoints {\bf Brownian beads}.

127: 

128: For a given cutpoint, the complement of the past path has one

129: infinite connected component. This can be mapped back to the half

130: plane via a conformal homeomorphism. It is convenient to normalize

131: this map so that it has derivative 1 at $\infty$ and takes the

132: cutpoint to 0; we call this map the {\bf conformal shift}.

133:  %

134: \begin{figure}

135: \label{f2} \centering

136: \includegraphics[height=3in]{bbf1}

137: \caption{Brownian excursion in $\HH$ and its cutpoints}

138: \end{figure}

139:  %

140: Our first goal is to show that Brownian beads are independent

141: of each other after a conformal transformation. More precisely,

142: for a cutpoint $g$ let $f$ denote the corresponding conformal

143: shift and let $\beta$ the bead starting at $g$ (if any).

144: \begin{theorem}\lb{ppp}

145: There exists a local time $\lambda$ supported on cutpoints so that

146:  $(f_{\lambda}(\beta_{\lambda}), \lambda \ge 0)$ is a Poisson

147: point process.

148: \end{theorem}

149: Here $f_\lambda$ applied to a path maps the image of the path

150: according to the conformal map, and changes time-parameterization

151: according to local Brownian scaling (Section \ref{s.conformal}).

152: The intensity measure of the Poisson point process is a

153: $\sigma$-finite measure on paths conditioned to start without

154: cutpoints; we call this measure the bead process. In Section

155: \ref{s.properties} we prove a Markov property for the bead process

156: and show that it has finite lifetime.

157: 

158: Theorem \ref{ppp} may be thought of as a two-dimensional analogy

159: of It\^o's theorem about Brownian excursions. Recall that

160: excursions of 1-dimensional Brownian motion are the segments of

161: the path between two consecutive visits to 0. For a zero $g'$ let

162: $\beta'$ denote the excursion starting at $g'$, and let $f'$

163: denote the map that shifts the future of the path by $-g'$.

164: \begin{theorem}[It\^o]

165: There exists a local time $\lambda'$ on zeros so that

166: $(f'_{\lambda'}(\beta'_{\lambda'}), \lambda' \ge 0)$ is a Poisson

167: point process.

168: \end{theorem}

169: The local time $\lambda'$ at zero has the nice property that the

170: $(g'(\lambda'),\lambda'\ge 0)$ process, or in other words the

171: total duration of excursions up to the zero $g'$ is a stable

172: subordinator of index $1/2$. It turns out that the analogous

173: statement holds in our setting. Let $a(t)$ denote the half-plane

174: capacity of the path up to time $t$; half-plane capacity is a way

175: to measure the size of subsets of $\HH$ (see Section

176: \ref{s.conformal}). Then

177: \begin{theorem}

178: The process $(a\circ g(\lambda), \lambda \ge 0)$ is a stable

179: subordinator of index $\alpha$.

180: \end{theorem}

181: Another property of beads which is shared by  It\^o excursions is

182: that the process $(\beta_\lambda,\,\lambda\ge 0)$ determines

183: $(B(t),\,t\ge0)$ (see Remark \ref{determine}).

184: 

185: In Section \ref{s.exponent} we show that the index $\alpha$ can be

186: expressed as an exponent.

187: 

188: \begin{theorem}\lb{exponent} We have $\pr(\mbox{B has no cuttime in }(1,t))

189: = t^{-\alpha + o(1)}.$

190: \end{theorem}

191: 

192: Finally, we compute $\alpha$.

193:  \newcommand{\indexhalfthm}

194:  {The Brownian bead index $\alpha$ equals $1/2$.}

195:  \begin{theorem}\lb{indexhalf}

196:  \indexhalfthm

197:  \end{theorem}

198: Although the $1/2$ here is equal to the index for It\^o's

199: excursions, it is a coincidence. Neither is it a consequence of

200: Brownian scaling. Unlike the proof of most exponents, this theorem

201: does not rely on SLE processes (nor exponents that use these

202: processes). The proof uses beads, not the exponent representation;

203: I am grateful to W. Werner with whom I discussed ideas of this

204: kind. A related exponent governing the Hausdorff dimension of

205: cuttimes for $\mbox{SLE}_6$ was computed by \cite{beffara}, and it

206: seems possible to derive the value of $\alpha$ directly from his

207: results. \cite{math.PR/0302115} also computed an exponent related

208: to $\alpha$ using SLE$(\kappa,\rho)$ processes (in a paper with

209: beautiful ideas and computations but few rigorous proofs). The

210: advantage of the proof here is that it is more elementary and

211: conceptual; it relies only on the special intersection exponent

212: $\xi(2,1)$, whose value was determined by \cite{lawler95}. His

213: proof covers even the 3-dimensional case.

214: 

215: In the proof, we will use the following fact about Brownian

216: excursion. Let $A$ be a compact subset of $\overline \HH\setminus

217: \{0\}$ so that the sets $\RR \cup A$ and $\HH\setminus A$ are

218: connected. Let $f:\HH\setminus A\to\HH$ be a conformal

219: homeomorphism fixing $\infty$ with $f'(\infty)=1$.

220: \begin{proposition}\lb{mainhit}

221: $\quad\pr(B\mbox{ avoids } A) =f'(0)$.

222: \end{proposition}

223: 

224: Further interesting properties of Brownian motion have recently

225: been found using the fact that its outer boundary has the same

226: distribution as (the outer boundary of) some SLE processes, see

227: \cite*{lwsrest} and the references therein. Some of methods used

228: in this paper appear independently in the recent literature; in

229: particular, see \cite{dubedat} and \cite{loopsoup}.

230: 

231: Many of the arguments of this paper carry over to ordinary

232: Brownian motion, more precisely, Brownian excursion from a

233: boundary point to an interior point of a domain. The analysis

234: there is a bit more difficult since there is no scale invariance;

235: we do not follow this avenue here.

236: 

237: In Section \ref{properties} we discuss some simple properties of

238: Brownian excursions in planar domains that we will need later. We

239: also show Proposition \ref{mainhit} there. In Section \ref{s.ctf}

240: we introduce the cuttime filtration and prove a Markov property

241: with respect to this filtration. Section \ref{s.conformal} reviews

242: the facts we need from conformal geometry, and introduces some

243: basic semigroups of paths. In Section \ref{s.independent} we prove

244: the Poisson point process decomposition of Theorem \ref{ppp}, and

245: we define the bead process, a Brownian excursion in the half plane

246: conditioned to start without cutpoints. In the next section we

247: show that this special process does not need infinite time to

248: start off, a fact that is not clear from the definition. In the

249: last two sections, \ref{s.exponent} and \ref{s.index}, we prove

250: Theorems \ref{exponent} and \ref{indexhalf}; we also give some

251: open questions and conjectures.

252: 

253: \section{Properties of Brownian excursion}\lb{properties}

254: 

255: Let $D$ be a {\bf regular domain}, i.e. a simply connected open

256: subset of the plane whose boundary is locally connected. Let

257: $\partial D$ denote the Caratheodory boundary, i.e. the set of

258: prime ends of $D$.

259: 

260: For $\{a\},Z\subset D\cup \partial D$, and $a\not=\infty$, let

261: $\BE(a,Z,D)$ denote Brownian motion started at $a$ and conditioned

262: to hit $Z$ no later than $\partial D$. If this event has positive

263: probability, then this definition is precise; otherwise it can be

264: made precise by considering $h$-processes or by taking a limit. A

265: special case is $\BE(0,\infty,\HH)$, the half-plane excursion,

266: which we will often abbreviate $\BE$. The coordinates of $\BE$ are

267: Brownian motion and an independent 3-dimensional Bessel process.

268: 

269: The two most important properties of $B\sim\BE(a,Z,D)$ are

270: restriction and conformal invariance. Restriction says that if

271: $D'\subset D$, then on the event that $B$ stays in $D'$,  it

272: has distribution $\BE(a,Z',D')$, where $Z'=Z\cap(D'\cup

273: \partial D')$. Conformal invariance says that if

274: $f:(a,Z,D)\to (a',Z',D')$ is a conformal homeomorphism between two

275: regular domains, then after a time change (\ref{timechange}) the

276: image of $B$ under $f$ has distribution $\BE(a',Z',D')$. The

277: restriction property is straightforward; conformal invariance has

278: been first proved by \cite{levy40}; the later development of

279: stochastic calculus makes it a simple exercise.

280: 

281: It is important for the previous paragraph that by Theorem 9.8

282: in the book of \cite{pommerenke}, conformal homeomorphisms from

283: the half plane $\HH$ or unit disk to regular domains extend

284: continuously to the boundary.

285: 

286: \begin{remark}\lb{curve} \rm

287: One way to get such domains is the following. Let $E$ denote the

288: image of a curve $\gamma:[0,1]\rightarrow \CC$, and let $D$ be a

289: connected component of $\CC\setminus E$. Since local connectivity

290: is preserved under continuous maps, $E$ is locally connected. The

291: proof of Theorem 9.8 in \cite{pommerenke}, part (ii) $\rightarrow$

292: (iii) carries through without changes to show that in fact the

293: boundary of $D$ is also locally connected.

294: \end{remark}

295: 

296: The strong Markov property says that the future of $B$ after a

297: stopping time $\tau$ given $\FF_\tau$ has distribution

298: $\BE(B(\tau),Z,D)$.

299: 

300: The following theorem is due to \cite{cm83} (see

301: \cite{MR86d:60088} for a simpler proof).

302: 

303: \begin{theorem}\label{cm}

304: There exists a universal constant $c$ so that the lifetime $\tau$

305: of a path distributed $\BE(a,Z,D)$ has $\ev\tau < c\;$area$(D)$.

306: \end{theorem}

307: 

308: We say that a sequence $a_n\in D$ {\bf converges to $a\in \partial

309: D$ along a path} if there is a continuous curve $\gamma:(0,1)\to

310: D$ so that $a_n=\gamma(t_n)$ for some sequence $t_n$, and $\gamma$

311: is contained in the equivalence class that defines the prime end

312: $a$.

313: 

314: In order to define a distance, we may extend paths defined on a

315: finite interval $[0,t]$ to be constant on $(-\infty,0]$ and

316: $[t,\infty)$. Let

317: \begin{eqnarray*}

318: \dist_0(\pi_1,\pi_2)&=&\sup_{t\in \RR}\|\pi_1(t)-\pi_2(t)\|,\\

319: \dist(\pi_1,\pi_2)&=&\inf_{\eps \in

320: \RR}\left(|\eps|+\dist_0(\theta_\eps\pi_1,\pi_2)\right).

321: \end{eqnarray*}

322: where $\theta_\eps$ denotes time shift. Let ``$\Rightarrow$''

323: denote convergence in distribution (in the case of paths with

324: respect to the metric ``$\dist$'').

325: 

326: \begin{proposition}\lb{kezdet}

327: If $a_n\to a\in \partial D$ along a path, then $\BE(a_n,z,D)

328: \Rightarrow \BE(a,z,D).$

329: \end{proposition}

330: 

331: \newcommand{\cdist}{\mbox{cdist}}

332: \begin{proof}

333: Consider the conformal homeomorphism $g:\HH \rightarrow D$ which

334: maps $0$ to the prime end $a$ (as determined by the path of

335: convergence of $a_n$). Since the domain $D$ is regular, we have $

336: g^{-1}(a_n)\rightarrow 0 $. Let

337: $$N_\eps=\{g(z)\ : \ z\in \HH, |z|<\eps\}.$$

338: Since the map $g$ is uniformly continuous in a neighborhood of

339: $0$, it follows that diam$(N_\eps)\rightarrow 0$ as $\eps \to 0$.

340: 

341: If $a_n\in N_\eps$, then let $p_{n,\eps}$ be the probability that

342: independent paths with distributions $\BE(a_n,Z,D)$ and

343: $\BE(a,Z,D)$ do not intersect before their respective hitting

344: times $\tau_{n,\eps}$, $\tau_\eps$ of $N_\eps$. By conformal

345: invariance, this probability can be computed in the half-plane

346: image. Then it is well known that there exists $c,\gamma>0$ so

347: that

348: $$

349: p_{n,\eps}:=\| \mu_{n,\eps} - \mu_{\eps}\|_{TV} <

350: c(|g^{-1}(a)-g^{-1}(a_n)|/\eps)^\gamma.

351: $$

352: Indeed, one can consider times of hitting concentric circles of

353: radii $2^{-k}$. Between such times the probability that two

354: Brownian excursions intersect is bounded away from $0$. Thus

355: $p_{n,\eps}\rightarrow 0$ for $\eps$ fixed and $n\rightarrow

356: \infty$.

357: 

358: Let $B$ have distribution $\BE(a,z,D)$, and define $B_n$ as having

359: distribution $\BE(a_n,z,D)$ and coupled to $B$ at its first

360: hitting time of the path of $B$. The event $C_n$ that this can be

361: done before either process has left $N_\eps$ has probability

362: $1-p_{n,\eps}$.

363: 

364: Let $A_n$ be the event that $|\tau_{\infty,\eps}-\tau_{n,

365: \eps}|<\delta$. By Theorem \ref{cm} and Markov's inequality,

366: $\pr(A_n^c)<c_2\; $area$(N_{\eps})/\delta$, for some universal

367: constant $c_2$. Now if $A_n$ and $C_n$ hold, and $\eps<\delta$,

368: then $\dist(B,B_n)<3\delta$. Therefore

369: $$

370: \pr(\dist(B,B_n)> 3 \delta) \le c_2\;\mbox{area}(N_{\eps})/\delta

371: + p_{n,\eps} + \one(\delta\le \eps)

372: $$

373: and the right hand side can be made arbitrarily small by first

374: picking $\eps$ then letting  $n\rightarrow \infty$. It follows

375: that $B_n$ converges to $B$ in probability and so in distribution.

376: \end{proof}

377: 

378: \begin{corollary}\label{01H}

379: Let $Z_n\rightarrow {1}$ be sets of reals, and let $a_n\in \HH$,

380: $a_n \rightarrow 0$. Then $$\BE(a_n,Z_n,\HH) \Rightarrow

381: \BE(0,1,\HH).$$

382: \end{corollary}

383: 

384: \begin{proof}

385: It suffices to prove this for the case when each $Z_n=\{z_n\}$ is

386: a single point, since the general case is a mixture of such

387: processes. Consider the rescaled versions $\BE(a_n/z_n,1,\HH)$ of

388: $\BE(a_n,z_n,\HH)$. These converge to $\BE(0,1,\HH)$ by the lemma,

389: and since the scaling factor converges to 1, so do the original

390: processes.

391: \end{proof}

392: 

393: Let $\BE(a,z,D,K)$ denote the distribution of $\BE(a,z,D)$

394: conditioned to hit $K$. Say a sequence of sets $A_n\subset D$ {\bf

395: converges to $a$ along a path} if every sequence of points $a_n\in

396: A_n$ does.

397: 

398: \begin{lemma}\label{slimhit}

399: Let $D$ be a regular domain, and consider regular subdomains $D_n

400: = D \setminus K_n$, where the $K_n \subset D$ are relatively

401: closed. Let $L_n \subset D_n$ closed, and let $a_n\in D_n$.

402: Suppose that

403: $$M_n:=K_n\cup L_n \cup \{a_n\} \rightarrow \{a\}\subset

404: \partial D  \quad \mbox{along a path}.$$ Then

405: $$\BE(a_n,z,D_n,L_n) \cd

406: \BE(a,z,D).$$

407: \end{lemma}

408: 

409: \begin{proof}

410: Let $W_n$ have distribution $\BE(a_n,z,D_n,L_n)$. The slightly

411: difficult part is to construct relatively closed sets $S_n

412: \rightarrow \{a\}$ separating $z$ from $M_n$ in $D$ so that

413: %

414:  \bel{e1}

415:  \pr(W_n \mbox{\ hits\ } S_n \mbox{ before }L_n)\rightarrow 0,

416:  \ee

417:  \vskip-1em

418:  \bel{e2}

419:  \sup_{s\in S_n} \pr(\BE(s,z,D) \mbox{ hits } K_n) \rightarrow 0.

420:  \ee

421: 

422: To complete the proof from here, let $\tau_n$ be the hitting time

423: of $S_n$ for $W_n$. Since $S_n \rightarrow \{a\}$, we have $\tau_n

424: \rightarrow 0$. If we condition $W_n$ to have done its duty of

425: hitting $L_n$ by time $\tau_n$, then $\{W_n(\tau_n+t),\ t\ge 0\}$

426: has the same distribution as $\BE(W(\tau_n),z,D_n)$. This, in turn

427: has the same distribution as $\BE(W(\tau_n),z,D)$ conditioned not

428: to hit $K_n$. In both cases, we are conditioning on events whose

429: probabilities converge to 1. Also, $\BE(W(\tau_n),z,D)$ is a

430: mixture of processes starting from points of $S_n$, so by

431: Proposition \ref{kezdet} it converges to $\BE(a,z,D)$. It follows

432: that the distribution of $W_n$ converges to $\BE(a,z,D)$, as

433: required.

434: 

435: Now we proceed to find the sets $S_n$. By conformal invariance

436: of the probabilities involved, it suffices to do this in the

437: upper half plane with $a=0,\ z=\infty$. Here we use that our

438: domain has a curve boundary, so if in $\HH$ we have $S_n

439: \rightarrow 0$, then for its image under the conformal

440: homeomorphism $(0,\infty,\HH) \to (a,z,D)$ we have $S'_n

441: \rightarrow a$.

442: 

443: Let $c_n\rightarrow \infty$ slow enough that we still have $c_n

444: M_n \rightarrow {0}$; if we construct $S_n$ that are bounded for

445: this  rescaled problem, then scaling back will make $S_n

446: \rightarrow {0}$. So it suffices to construct bounded $S_n$. We

447: consider the uniformizing map $g_n$ which takes $D_n$ to the half

448: plane $\HH$. Assume that $g_n$ has hydrodynamic normalization, and

449: extend it to the boundary of $D_n$.  The half-plane version of the

450: Caratheodory Kernel Theorem (see Theorem 1.8 in \cite{pommerenke})

451: implies that $g_n(z)\rightarrow z$ uniformly where it is defined.

452: So if we $L_n'=g_n(L_n)$, $a_n'=g_n(a_n)$ then

453: $L'_n\cup\{a'_n\}\rightarrow \{0\}$.

454: 

455: 

456: Let $S$ be the unit semicircle in $\HH$, and use the shorthand

457: $B_n=\BE(a'_n,\infty,\HH)$. We will show that

458:  \bel{e5}

459:  \pr(B_n \mbox{ hits } S \mbox{ before } L'_n\ |\ B_n \mbox{ hits }L'_n

460: )\rightarrow 0,

461:  \ee

462: \vskip-1em

463:  \bel{e6}

464:  \sup_{s\in S}\pr (\BE(s,\infty,\HH) \mbox{\ hits\ } L'_n) \rightarrow 0.

465:  \ee

466: Once we have this, we take $S_n= g_n^{-1}(S)$ so by uniform

467: convergence, $S_n \rightarrow S$. Thus  $\{S_n\}$ is bounded, and

468: $(\ref{e1}, \ref{e2})$ follow from conformal invariance.

469: 

470: The second statement (\ref{e6}) is obvious, and the first follows

471: since we can easily find semicircles $R_n \rightarrow \{0\}$ so

472: that

473:  \bel{e7}

474:  \sup_{w\in R_n}

475:  \pr(\BE(w,\infty,\HH)\mbox{\ hits\ } L'_n) <

476:  \pr(\BE(a'_n,\infty,\HH)\mbox{\ hits\ } L'_n).

477:  \ee

478: The left hand side of (\ref{e5}) can be written as

479:  \bel{e3}

480:  \pr(\mbox{$B_n$ hits $S$ before $L'_n$ and hits $L'_n$})

481:  \ /\ \pr(\mbox{$B_n$ hits }L'_n).

482:  \ee

483:  Using the Markov property and the topology of the setup the

484: numerator can be bounded above by

485: \begin{equation} \label{e4}

486: \sup_{w \in S} \pr(\BE(w,\infty,\HH)\mbox{ hits }R_n)

487:   \sup_{w\in R_n}\pr(\BE(w,\infty,\HH)\mbox{ hits }L'_n)

488: \end{equation}

489: and so by (\ref{e7}), expression (\ref{e3}) is bounded above by

490: the first factor of (\ref{e4}), which converges to 0, proving

491: (\ref{e5}).

492: \end{proof}

493: 

494: 

495: \begin{lemma}\label{fathit}

496: Let $D$ be a regular domain, and let $K\subset D$ be a relatively

497: compact connected set containing at least two points. Let $z\in

498: \partial D$. Suppose that $K_n\subset D$ are relatively compact

499: and converge to $K$. Suppose that $a_n \rightarrow a$. Then

500: $$\BE(a,z,D,K_n)\cd \BE(a_n,z,D,K).$$

501: \end{lemma}

502: 

503: \begin{proof}

504: Let $g$ be a bounded continuous test function for paths, and let

505: $B_n,\,B$ be distributed as $\BE(a_n,z,D)$ and $\BE(a,z,D)$

506: respectively.

507: 

508: Note that for $\BE(a,z,D)$-almost all paths $\pi$ (1) the function

509: $f_n = \one(\pi\ hits\ K_n)$  converges to $f = \one(\pi\ hits\

510: K)$ and (2) $f$ is continuous at $\pi$. Thus by Proposition

511: \ref{kezdet} we have

512: \begin{eqnarray*}\ev(f_n(B_n)g(B_n)) &\rightarrow& \ev(f(B)g(B)),

513: \\

514: \ev f_n(B_n) &\rightarrow& \ev f(B).

515: \end{eqnarray*}

516: Note that the hypothesis of the lemma ensures that $\ev f(B)$ is

517: positive. Taking the ratio of the previous two limit statements we

518: get the desired result.

519: \end{proof}

520: 

521: \subsection*{Hitting probabilities}

522: 

523: Let $D$ be a domain, and let $a,z$ be points on $\partial D

524: \setminus A$ in the neighborhood of which the boundary is a

525: differentiable curve. Let $A$ be a hull in $\overline D$ not

526: containing $a,z$. We show the following equivalent version of

527: Proposition \ref{mainhit}.

528: 

529: \begin{proposition}\lb{hit}

530: Let $f$ be a conformal homeomorphism that takes $D\setminus A$ to

531: $D$ and fixes $a,z$. Then

532: $$

533: \pr(\BE(a,z,D)\mbox{ avoids }A) = f'(a)f'(z).

534: $$

535: \end{proposition}

536: 

537: \begin{proof}

538: Because of conformal invariance, it is sufficient to prove this in

539: the half-plane $\HH $ with $a=0$, $z=1$.

540: 

541: The indicator function of the event above is continuous at almost

542: every path with respect to the excursion measure. Let $a_n=i/n$,

543: and $Z_n=(1,1+1/n)\subset \RR$. It suffices to show that as

544: $n\rightarrow \infty$ we have

545: \begin{equation}\label{ff1}

546: \pr(\BE(a_n,Z_n,\HH)\mbox{ avoids }A) \rightarrow f'(0)f'(1)

547: \end{equation}

548: since the measures on the left converge to the excursion measure

549: by Corollary \ref{01H}.

550: \begin{eqnarray}

551: \pr(\bm(a)\mbox{ hits }Z_n\mbox{  no later than }\partial

552: D)=h_{1/n}(i/n) = \nu n^{-2} + o (n^{-2}) \label{hit1}

553: \end{eqnarray}

554: Here $h_\eps$ is the harmonic function determined by the boundary

555: conditions $1$ on $(1,1+\eps)$ and $0$ elsewhere, and $\nu$ is

556: some fixed constant. The functions $h$ can be easily given

557: explicitly, but the formulas are unimportant, all we need is the

558: intuitive fact that as $\eps,z \rightarrow 0$ we have

559: $$h_\eps(z) = \nu\eps\Im z + o(\eps|z|)

560: $$

561: giving the second equality in (\ref{hit1}). Similarly, after a

562: conformal transformation by $f$

563: \begin{equation}\label{hit2}

564: \pr(\bm(a_n)\mbox{ hits }Z_n \mbox{ no later than }A\cup\partial

565: D)

566: \end{equation}

567: becomes

568: \begin{eqnarray}

569: \pr(\bm(f(a_n))\mbox{ hits }f(Z_n) \mbox{ no later than

570: }f(A\cup\partial \HH)) \quad  \nonumber

571: \\=h_{f(1+\eps)-1}(f(i\eps))  = f'(0)f'(1)\nu

572: n^{-2} + o(n^{-2}) \nonumber

573: \end{eqnarray}

574: but the ratio of (\ref{hit2}) and (\ref{hit1}) gives (\ref{ff1}),

575: proving the claim.

576: \end{proof}

577: 

578: \section{The cuttime filtration} \lb{s.ctf}

579: 

580: Consider the excursion $\{B_t\}$ with distribution $\BE(a,z,D)$

581: where $D$ is a regular domain, $a,z\in \partial D$. Let $G$ denote

582: the set of {\bf cuttimes}, that is, the set of times for which

583: images of the future and past are disjoint; \cite{burdzy89} showed

584: that such times exist with probability 1. It is easy to check that

585: the set $G$ is measurable and it is a closed set a.s.

586: 

587: It is clear that one cannot decide what the cuttimes are up to

588: time $t$ by only looking at the past of the process $\{B_t\}$. $G

589: \cap [0,t]$ is therefore not measurable with respect to the

590: standard filtration ${\mathcal F}_t$ of $\{B_t\}$. We therefore

591: introduce the {\bf cuttime filtration} $\mathcal G_t$ generated by

592: ${\mathcal F}_t$ and $G \cap [0,t]$.

593: 

594: This enlargement of filtration may look large, but in fact it is

595: not in the following sense. Let $\Ghat_t$ denote the set of

596: cuttimes of the process $\{B_t\}$ restricted to the interval

597: $[0,t]$. If $\tau$ is the earliest time for which the set of

598: points $B[0,\tau]$ is revisited after time $t$, then we clearly

599: have

600: $$

601: G \cap [0,t] =  \Ghat_t \cap [0,\tau].

602: $$

603: Since $\Ghat_t$ has Hausdorff dimension less than one

604: \citep{lawler96}, and the distribution of $B_\tau$ is absolutely

605: continuous with respect to harmonic measure on $B[0,t]$, we have

606: by Makarov's theorem that  $\tau \notin \Ghat_t$ ${\mathcal

607: F}_t$-a.s. In particular, $G\cap[0,t]$ is determined by the

608: connected component $(g_t,g_t')$ of $[0,t] \setminus \Ghat_t$

609: containing $\tau$. The left endpoint of this interval, $g_t$, is

610: the last cuttime up to time $t$. Then given ${\mathcal F}_t$,

611: $g_t$ determines $\mathcal G_t$; its distribution  is concentrated

612: on the countable set of points in $\Ghat_t$ that are isolated from

613: the right.

614: 

615: Let $\rho B_s$ denote the connected component of $D\setminus

616: B[0,s]$ having $z$ as an interior or boundary point. By Remark

617: \ref{curve} if $D$ is a regular domain, then so is $\rho B_s$ for

618: every $s$. Let $\BE(a,z,D,K)$ denote the distribution of

619: $\BE(a,z,D)$ conditioned to hit $K$.

620: 

621: \begin{proposition}[Markov property for cuttime filtration]

622: \lb{mp} Let $\tau$ be a ${\mathcal G}_t$-stopping time. Then the

623: distribution of $\{B_{\tau+t}, t\ge 0\}$ given $\mathcal G_\tau$

624: is $\BE(B_t,z,\rho B_{g_t}, B[g_t,g_t'])$.

625: \end{proposition}

626: 

627: Note that, while for fixed time $\tau$, $B_{\tau}$ is an interior

628: point of $\rho B_{g_\tau}$ with probability one, there exists

629: stopping times for which $g_\tau=\tau$ a.s. An example is the

630: first cuttime after time 1. In these cases the ``hitting''

631: condition of the proposition holds trivially.

632: 

633: \begin{proofof}{Proposition \ref{mp}} \

634: 

635: Recall that $\GG_t$ is generated by $ \FF_t$ and $g_t$. Since the

636: distribution of $\{B_{t+s},s \ge 0\}$  given $\FF_t$ is

637: $\BE(B_t,z,D)$ by the Markov property, further conditioning on the

638: value of $g_t$ means that the future of the process $\BE(B_t,z,D)$

639: has to stay in $\rho B_{g_{\tau}}$, and hit

640: $B[g_{\tau},g'_{\tau}]$. This completes the proof for fixed time

641: $t$. Denote $L_t$ the set $B[g_{t},g'_{t}]$, and denote $\ev_t$

642: the distribution conditioned as above, that is $\BE(B_t,z,\rho

643: B_t, L_t)$.

644: 

645: What we  showed is equivalent to the following. If $X$ is a

646: bounded random variable on paths, and the time shift operator

647: $\theta_t$ is defined as $ (\theta_t B)(s)=B(s+t)$, then

648: $$

649: \ev (X\circ \theta_t \ |\ \GG_t) = \ev_t X.

650: $$

651: Thus, if $\tau$ is a $\GG_t$-stopping time concentrated on a

652: discrete set of values $\mathcal T$, then

653:  \begin{eqnarray*}

654:  \ev (X\circ\theta_\tau\ |\ \GG_\tau)

655:  &=& \sum_{t\in \mathcal T} \ev (X\circ \theta_\tau\ \one(\tau=t) \ |\ \GG_\tau)

656:  \ =\ \sum_{t\in \mathcal T} \ev (X\circ \theta_t\ \one(\tau=t) \ |\ \GG_t)

657:  \\&=& \sum_{t\in \mathcal T} \one(\tau=t) \ev (X\circ \theta_t\ \

658:  |\

659:  \GG_t)

660:  \ =\  \sum_{t\in \mathcal T} \one(\tau=t) \ev_t X

661:  \ =\ \ev_\tau X,

662:  \end{eqnarray*}

663: which completes the proof for this case. For an arbitrary

664: $\GG_t$-stopping time $\tau$ we take a discrete approximation by

665: letting $\tau_n$ be the first element of $ 2^{-n}\Z$ which is at

666: least $\tau$. Since the decision to stop at time $\tau_n$ is made

667: by time $\tau$, we have

668: $$

669: \ev(X\circ \theta_{\tau_n} \ |\ \GG_{\tau_n}) = \ev(X\circ

670: \theta_{\tau_n} \ |\ \GG_\tau).

671: $$

672: Since $\theta_t$ is a.s. continuous as $t \downarrow 0$, and $X$

673: is continuous, we have $X \circ \theta_{\tau_n} \rightarrow

674: X\circ\theta_{\tau}$ almost surely, and by the bounded convergence

675: theorem

676: $$

677: \ev(X\circ\theta_{\tau_n} \ | \ \GG_\tau) \rightarrow \ev(X\circ

678: \theta_\tau \ |\ \GG_\tau).

679: $$

680: Therefore it suffices to prove that $\FF_\infty$-a.s.

681: $\ev_{\tau_n} X \rightarrow \ev_\tau X$, or

682:  \bel{cd}

683:  \BE(B_{\tau_n}, z, \rho B_{g_{\tau_n}},L_{\tau_n}) \cd

684:  \BE(B_{\tau},   z, \rho B_{g_\tau},L_{\tau}).

685:  \ee

686: First we fix the path $B$ and examine what happens to $g_{\tau_n}$

687: and $g'_{\tau_n}$ in the limit. The definition $g_\tau= \sup (G

688: \cap [0,\tau])$ implies that $g_\tau$ is increasing in $\tau$ and

689: continuous from the right. Assume that $\tau\not=g_\tau$. Then

690: $\tau$ lies in a connected component of $G$ whose left endpoint is

691: $g_\tau$, so for all large $n>N_1$ we must have $g_{\tau_n}=

692: g_\tau$.

693: 

694: Now assume also that $g'_\tau \not=\tau$. Then $B[g_\tau,

695: g_\tau']$ is disjoint from $B(g_\tau', \tau]$. Let $t$ be the

696: first time after $\tau$ that $B_t$ revisits the set

697: $B[g_\tau,g_\tau']$. Then for $\tau_n<t$ and $n>N_1$ we have

698: $g'_{\tau_n}=g'_\tau$. On the other hand, if $g'_\tau=\tau$, then

699: $ g'_{\tau_n}\le {\tau_n}$ implies that $g'_{\tau_n}\downarrow

700: g'_{\tau}$.

701: 

702: So whenever $\tau\not=g_\tau$, we have $\rho B_{g_{\tau_n}} = \rho

703: B_{g_\tau}$ eventually, $L_{\tau_n} \rightarrow L_{\tau}$, and the

704: latter contains at least two points. By Lemma \ref{fathit} we get

705:  $$

706:  \BE(B_{\tau_n}, z, \rho B_{g_\tau},L_{\tau_n}) \cd

707:  \BE(B_{\tau},   z, \rho B_{g_\tau},L_{\tau}),

708:  $$

709: which implies (\ref{cd}).

710: 

711: If $g_\tau=\tau$, then set $K_n:=B[g_{\tau},g_{\tau_n}]$. Note

712: that $K_n \cup L_{\tau_n} \cup \{B_{\tau_n}\} \subset B[\tau,

713: \tau_n]$ converges to $\{B_\tau\}$ along a path (i.e. a sub-path

714: of $B_t$). Therefore Lemma \ref{slimhit} yields (\ref{cd}).

715: \end{proofof}

716: 

717: \section{Conformal maps, paths, semigroups} \lb{s.conformal}

718: 

719: We begin this section with some notation and standard facts about

720: complex geometry. Most can be found in \cite{pommerenke},

721: \cite{lawlersle}, and best of all the upcoming book

722: \cite{lawlerbook}, which reviews complex geometry with an eye

723: towards applications in stochastic processes.

724: 

725: If $A\subset \HH$, let  $h_y(A)$ be harmonic measure, or the

726: hitting distribution of Brownian motion started at $y$ and stopped

727: at $A\cup \RR$. Let

728: $$h_\infty(A)=\lim_{y\rightarrow\infty} y\cdot

729: h_y(A)$$ Define

730: \begin{eqnarray}\label{capz}

731:  \capz(A)&=&h_\infty(A),\\

732:  \label{capo}

733:  \capo(A)&=&\int_{\overline A} \Im(z) h_\infty(dz).

734: \end{eqnarray}

735: If $D$ is a subset of $\HH$ so that $\HH\setminus D$ is bounded,

736: then the capacity $\capz$ in $D$ can be defined analogously and is

737: denoted $\capz(D,A)$. Conformal invariance of harmonic measure

738: implies that if $f$ is a conformal homeomorphism $D\rightarrow

739: \HH$ satisfying $f'(\infty)=1$, then $\capz(D,A)=\capz(f(A))$. The

740: quantity $\capo$ will be referred to as half-plane capacity.

741: 

742: If $A$ is bounded, $A \cup \RR$ is connected, and $f:\HH\setminus

743: A \to \HH$ is a conformal homeomorphism with hydrodynamic

744: normalization (i.e. $f(z)-z\to 0$ as $z\to\infty$), then we have

745:  \bel{hyd}

746:  f(z)=z+\capo(A)/z+O(1/z^2)

747:  \ee

748: as $z\rightarrow \infty$. Such maps have the property that for all

749: $z\in \HH$

750:  \bel{shiftdown}

751:   \Im f(z)\le \Im(z).

752:  \ee

753: Recall the following standard fact about complex geometry. There

754: exist a constant $c$ so that if $A\subset \overline \HH$ is

755: connected, then

756:  %

757:  \bel{capdia}

758:  c^{-1} \diam(A)\le\capz(A)\le c\,\diam(A),

759:  \ee

760:  %

761: if $A\cup \RR$ is connected and $x+iy\in A$, then

762:  %

763:  \bel{capim}

764:  y^2/4 \le \capo(A).

765:  \ee

766: Also, if $A,A'$ are disjoint, then we have

767:  %

768:  \bel{suba}

769:  \capo(A\cup A')\le  \capo(A)+\capo(A')

770:  \ee

771:  %

772: This implies that if $A_1,A_2,A$ are disjoint, with $A_1\cup R$

773: and $\HH\setminus A_1$ connected, then

774:  %

775:  \bel{concap}

776: \capo(A_1 \cup A_2 \cup A)-\capo(A_1 \cup A) \le \capo(A_1\cup

777: A_2)-\capo(A_1)

778:  \ee

779:  %

780: We apply the conformal map $f:(\HH\setminus A_1,0,\infty)\to

781: (\HH\setminus A_1,0,\infty)$; then by additivity of $\capo$ the

782: desired inequality transforms to $\capo(f(A_2\cup A))-\capo(f(A))

783: \le \capo(f(A_2))$, which follows from \re{suba}.

784: 

785: Let $\Phi$ denote the set of paths $\pi:[0,\tau]\rightarrow

786: \overline \HH$ which intersect the real line only at time $0$ and

787: $\pi(\tau)$ is in the boundary of $\rho\varphi$, the infinite

788: connected component of the complement of the image $\pi[0,\tau]$.

789: 

790: To each $\pi \in \Phi$ we can associate the unique conformal

791: homeomorphism $f_\pi:\HH \rightarrow \rho_\pi$ with

792: hydrodynamic normalization. which fixes $\infty$, has

793: derivative $1$ there, and extends to the boundary $\RR$

794: continuously (see Remark \ref{curve}) and maps $0$ to $\pi(t)$.

795: Such a map has expansion

796:  \bel{shiftexp}

797:  f(z)=z + a_0 + az^{-1} +

798: O(z^{-2})

799:  \ee

800: at $\infty$. The coefficient $a=\capo(\pi)$ behaves additively

801: under compositions of maps. It has the scaling property

802:  \bel{scaling}

803:  a(r \pi)=r^2 a(\pi).

804:  \ee

805: Using the conformal maps, it is possible to compose two paths in

806: $\Phi$. Let $\pi\circ \pi'$ equal $\pi$ on $[0, \ell_\pi]$, and

807: equal the $f_\pi$-transform of $\pi'$ (see the definition below)

808: time-shifted by $\ell_\pi$ for the rest of the time interval.

809: 

810: Thus $\Phi$ is a semigroup, and the map $a$ is a semigroup

811: homomorphism $\Phi \rightarrow [0,\infty)$.

812: 

813: \begin{definition} \rm Let $\pi$ be a path in a regular domain $D_1$

814: (possibly starting and ending at $\partial D_1$), and let $f$

815: be a conformal homeomorphism $D_1 \rightarrow D_2$. If the

816: integral

817:  %

818:  \bel{timechange}

819: s(t)=\int_0^t |f'\circ \pi(\tau)|^2 d\tau

820:  \ee

821:  %

822: is finite for each $t$, then let $t(s)$ be the inverse of $s(t)$;

823: the function $f\circ \pi\circ t (s)$ is called the {\bf

824: $f$-transform} of $\pi(t)$.

825: \end{definition}

826: 

827: \section{Independent increments and a local time for beads}

828: \lb{s.independent}

829: 

830: Consider the process $B\sim\BE$, and let $G$ denote the set of

831: cuttimes. If $B(t)$ is on the boundary of $\rho B_t$, let $f_t$

832: denote the normalized conformal homeomorphism mapping $(\rho

833: B_t,B(t), \infty)$ to $(\HH,0,\infty)$. In this section, we

834: explore the following consequence of Proposition \ref{mp} and the

835: conformal invariance of Brownian excursion.

836: 

837: \begin{corollary}[Independent increments at cuttimes]

838: \label{iip} Let $\tau$ be a ${\mathcal G}_t$-stopping time

839: supported on the set of cuttimes $G$. Then the $f_\tau$-map of the

840: future has distribution $\BE$.

841: \end{corollary}

842: 

843: By a minor abuse of notation,  let $a(t)$ denote the half-plane

844: capacity of $B[0,t]$.

845: 

846: \begin{lemma} The set $a(G)$ is a closed regenerative set

847: with a scale-invariant distribution.

848: \end{lemma}

849: Recall that a random set $S$ has scale-invariant distribution for

850: every $c>0$ the sets $cS$ are identically distributed. It is

851: called regenerative if for every $\{\sigma[S \cap [0,t]];\; t\ge

852: 0\}$-stopping time $\nu$ which satisfies $\nu\in S$ a.s. the

853: translated set $(S-\tau) \cap [0,\infty)$ has the same

854: distribution as $S$.

855: 

856: \begin{proof}

857: Closedness follows from the continuity of half-plane capacity and

858: since the set of cuttimes is closed. Scale-invariance follows from

859: the scale-invariance of Brownian excursion and the scaling

860: property $\re{scaling}$ of the half-plane capacity $a$.

861: 

862: To check the regenerative property, assume that the stopping time

863: $\nu$ is as above. Then $\tau=a^{-1}(\nu)$ is a stopping time for

864: the cutpoint filtration $\GG$, and $\tau=a^{-1}(\nu)$ is supported

865: on cutpoints. Therefore by Corollary \ref{iip} the distribution of

866: the $f_\tau$-mapping of the future of $B$ after $\tau$ has

867: distribution $\BE$.

868: 

869: In particular, since $a$ behaves additively under conformal

870: homeomorphisms, the distribution of $(a(G)-\nu)\cap[0,\infty)$

871: given the past of the excursion is the same as the distribution

872: of $a(G)$. Thus $a(G)$ is a regenerative set, as required.

873: \end{proof}

874: 

875: By a result of \cite{kingman73}, the above implies that $a(G)$ is

876: the image of a stable subordinator. More precisely, there exists a

877: nondecreasing random function $(\lambda(a),\, a \ge 0)$ adapted to

878: the filtration $(\sigma(a(G) \cap [0,a]),\, a \ge 0)$ so that

879: $\lambda$ increases exactly on the set $a(G)$ its right-continuous

880: inverse $(a_\lambda,\, \lambda\ge 0)$ is a stable subordinator

881: with index $\alpha\in (0,1)$. The normalization can be chosen so

882: that the L\'evy measure (the intensity of the Poisson point

883: process of jumps) assigns mass $y^{-\alpha}$ to the interval

884: $(y,\infty)$ for all $y\ge 0$. We call $\lambda$ the {\bf bead

885: (local) time}.

886: 

887: Let $B_{\rightarrow \lambda}$ denote the process $(B(t), 0\le t

888: \le t(\lambda))$ as an element of the semigroup $\Phi$. Recall

889: that a L\'evy process $X$ on a topological semigroup $\Phi$ is a

890: right continuous process with left limits with the property that

891: at any fixed time $t$, given the entire past $(X_s, 0\le s \le

892: t)$, the distribution of $X$ is the same as the past composed with

893: an independent copy of $X$. Since the inverse bead local times are

894: $\GG_t$-stopping times supported on $G$, Corollary \ref{iip}

895: immediately gives

896: 

897: \begin{proposition}$(B_{\rightarrow \lambda},\, \lambda\ge 0)$ is a

898: $\Phi$-valued L\'evy process.

899: \end{proposition}

900: 

901: Let $t(\lambda)$ be the time corresponding to local time

902: $\lambda$, i.e. the solution of $a(t(\lambda))=a_\lambda$.

903: Whenever the process $a_\lambda$ has a jump, the segment

904: $(B(t),\,t\in[t(\lambda-),t(\lambda+)])$ has no cutpoints, while

905: $t(\lambda-),\,t(\lambda+)$ are cuttimes. Let $\beta(\lambda)$

906: denote the $f$-mapping of this segment from $\rho B(t(\lambda-))$

907: to $\HH$. Then $\beta(\lambda)$ is an element of the semigroup

908: $\Phi$. When $a(\lambda)$ has no jump, we set $\beta(\lambda)$

909: equal some null state.

910: 

911: Note the following deterministic fact. For an interval

912: $I=[\lambda_1,\lambda_2]$ the set $\beta(I)$ equals the set

913: $\beta'(I-\lambda_1)$ for the process which is the

914: $f_{a(\lambda_1)}$-image of the original one. Since the inverse

915: bead local times are stopping times, the independent increment

916: property of Corollary \ref{iip} implies that for non-overlapping

917: intervals $I$, the sets $\beta(I)$ are independent. We have shown

918: 

919: \begin{proposition} $\beta$ is a

920: $\Phi$-valued Poisson point process.

921: \end{proposition}

922: 

923: The {\bf Brownian bead measure} ($\BB$) is the $\sigma$-finite

924: intensity measure of the Poisson point process $\beta$. In simple

925: terms, for a set of paths $A$ the measure $\BB(A)$ equals the

926: expected number of elements of $A$ among the beads $\beta[0,1]$.

927: 

928: \section{Properties of beads}\lb{s.properties}

929: 

930: The goal of this section is to establish some simple properties of

931: the {\bf bead process}, i.e. the measure $\BB$.

932: 

933: {\bf Scaling.} It follows from the scaling properties of $\BE$ and

934: half-plane capacity that for $c>0$ we have

935: $\BB(rA)=r^{-2a}\BB(A)$, where scaling a path by $c$ means scaling

936: in space by a factor $c$ and time accordingly. This implies that

937: the law $\BB$ can be decomposed as a product of measures on

938: ``shape space'' and ``size space''. Let $\beta$ be chosen

939: according to $\BB$ conditioned to have size $a$ at least $1$, and

940: rescale $\beta$ to have size 1. The resulting probability measure

941: $\nbb$, determines the shape of $\BB$, while the size is given

942: independently by the measure $d(y^{-\alpha})$.

943: 

944: We now check that the Markov property for cuttimes and conformal

945: invariance imply a Markov property of beads.

946: 

947: Note that the time of beads is only defined up to translation

948: (since the integral in the $f$-transform may not be finite). We

949: may pick a rule to set $t=t_1$ for some fixed number $t_1$ when

950: (and if) the bead grows to size 1 (which here is an arbitrary

951: positive number). Let $t_0$ be the starting time of the bead

952: (possibly $-\infty$), and let $A_t$ denote the image of the time

953: interval between $t_0$ and the first $t$-local cuttime after $t_0$

954: under $B$. Let $T\ge t_1$ be a stopping time with respect to the

955: canonical filtration generated by the past of the process.

956: 

957: \begin{lemma}[Markov property of beads]\lb{mpb}

958: Under the measure $\BB$, the process $(B(T+s),\, s\ge 0)$ has the

959: same distribution as a process $B'$ distributed as

960: $\BE(B(T),\infty, \HH,A_T)$ and stopped at a random time $\tau$.

961: Here $\tau$ is the first time that

962: $$(B[t_0,\tau]\cup B'[0,\tau))\cap B'(\tau,\infty] =0.$$

963: \end{lemma}

964:  %

965: \begin{proof}

966: It suffices to show this for the case $T=t_1$ (recall that the

967: bead starts at a time $t_0<t_1$). The general case follows from

968: first applying the lemma at $T=t_1$ and then using the Markov

969: property of Brownian excursion. More precisely, if $B'$ is as

970: given in the claim, and $T\ge t_1$ is a stopping time, then the

971: distribution of the future of $B$ after $T$ is just the

972: distribution of the future of $B'$ after $T$, and $B'\sim

973: \BE(B(T),\infty, \HH,A_T)$ stopped at $\tau$.

974: 

975: For the $T=t_1$ case note that $\BB$ conditioned to have size at

976: least $1$ is by definition the following. $\BE$ is run until the

977: first time $S$ that $a(S)-a(g'(S))=1$ where $g'(S), g(S)$ are the

978: global cuttimes immediately before and after time $S$. Then $\BB$

979: is the $f_{g'(S)}$-map of $B(g'(S)+s,\,0\le s \le g(S)-g'(S))$.

980: 

981: Therefore the statement of the lemma follows from the Markov

982: property of Proposition \ref{mp}, conformal invariance and the

983: fact that $f$-mappings preserve half-plane capacity.

984: \end{proof}

985: 

986: Note that an example of such a stopping time is the first time

987: when $\BB$ hits the line with imaginary part $y$. This will be

988: used in Lemma \ref{lifetime} to show that in fact beads have

989: finite lifetime, and can be started at $t_0=0$. Once this has been

990: done, it is straightforward to extend Lemma \ref{mpb} to arbitrary

991: stopping times $T>t_0$.

992: 

993: \begin{remark}[Beads determine the excursion]\lb{determine}\rm Another property shared by

994: beads and It\^o excursions is that the process $(\beta_\lambda,\,

995: \lambda \ge 0)$ determines the process $(B(t),\, t\ge 0)$. When

996: $B(t)$ is on the boundary of $B[0,t]$ (these are the so-called

997: pioneer points), the parameter $a_0$ in the conformal shift

998: \re{shiftexp} defines the ``horizontal'' location of $B(t)$. Given

999: $\beta$, it is straightforward to determine $a_0$ as a function of

1000: the half-plane capacity $a$ for each $a\ge 0$. This gives a

1001: L\"owner chain for the pioneer points of $B$ (see, for example,

1002: \cite{lawlersle} for definitions), and hence determines its outer

1003: boundary. Now assume that the outer boundaries of $B$ and $B'$

1004: agree, but they still differ in some way within a particular bead.

1005: Since $f$-mappings are one-to-one, then that bead has to be mapped

1006: by the same conformal shift to different points in the processes

1007: $\beta, \beta'$. This answers a question posed by an anonymous

1008: referee.

1009: \end{remark}

1010: 

1011: The proof of the following simple fact is left to the reader.

1012: 

1013: \begin{lemma}  \lb{time}

1014: Let $A$ be a nonempty subset of $\RR\times (0,1]\subset \HH$ so

1015: that $A\cup \RR$ is connected. Let $B$ have distribution

1016: $\BE(z,\infty, \HH)$ conditioned to hit $A$ and stopped when this

1017: happens. Let $T_1$ be the time $B$ spends in the strip with

1018: imaginary parts between $1$ and $2$. There exists absolute

1019: constants $c,\gamma$ so that for all $t>0$ we have

1020: $$

1021: \pr(T>t) < c e^{-\gamma t}.

1022: $$

1023: \end{lemma}

1024: 

1025: \begin{lemma}[Finite lifetime]\lb{lifetime}

1026: Let $T$ denote the lifetime of the process with distribution

1027: $\BB$. Then a.e. $T<\infty$.

1028: \end{lemma}

1029: 

1030: \begin{proof}

1031: Let $B$ be a bead conditioned to hit $\Im z=y$. By the Markov

1032: property, the future of $B$ after this time is just $\BE(z,\infty,

1033: \HH)$ conditioned to hit a certain subset of the past. Let

1034: $T(y,2y)$ denote the time $B$ spends with imaginary part in this

1035: interval. By Lemma $\ref{time}$ and scale invariance

1036: $$

1037: \pr(T(y,2y)>ty^2) < ce^{-\gamma t}

1038: $$

1039: and therefore, for the unconditioned measure

1040: $$

1041: \BB(T(y,2y)>ty^2) < y^{-\alpha}c'e^{-\gamma t}.

1042: $$

1043: setting $y_n=2^{-n}$, and $t_n=c_1 n$ the right hand side becomes

1044: $ 2^{\alpha n}2^{-c_1\gamma n} $ which is summable for an

1045: appropriate choice of $c_1$. By the first Borel-Cantelli lemma

1046: (which also holds for $\sigma$-finite measures) we get that $\BB$

1047: a.e. for all large $n$

1048: $$

1049: T(2^{-n},2^{1-n}) \le c n 2^{-2n}

1050: $$

1051: summing this we get that for some ``random'' constant $K$ and all

1052: $y\le 1$ \bel{timeineq} T(0,y)< K y^2|\mbox{log}\,y| \ee But the

1053: Markov property and the existence of cutpoints for large times

1054: implies that after hitting $\Im(z)=1$, $B$ will have finite

1055: lifetime $T'$ a.s. Therefore $T\le T(0,1)+T'$ is finite $\BB$-a.e.

1056: as required.

1057: \end{proof}

1058: 

1059: \section{The exponent giving the bead index }\lb{s.exponent}

1060: 

1061: The goal of this section is to identify the index $\alpha$ of the

1062: stable process driving Brownian beads as an exponent for a large

1063: deviation event. Let $A(t)$ denote the event that the half-plane

1064: excursion $B$ has no cuttime between times $1$ and $t$.

1065: 

1066: \newtheorem*{exponent.thm}{Theorem \ref{exponent}}

1067:  \begin{exponent.thm}

1068: For large $t$, we have $\pr A(t) = t^{-\alpha + o(1)}$.

1069: \end{exponent.thm}

1070: 

1071: \cite{math.PR/0302115} recently computed essentially the same

1072: exponent using generalized SLE processes, with the result

1073: $\alpha=1/2$. We will compute $\alpha$ directly by a simpler

1074: argument in the next section.

1075: 

1076: Let $(a(\lambda),\lambda \ge 0)$ be a stable subordinator with

1077: index $\alpha$. We will use the following simple fact. It follows

1078: directly from \cite{bertoin} page 76 Proposition 2.

1079: 

1080: \begin{fact}\label{bert} Let

1081: \[

1082: X =\min\left(a([0,\infty)) \cap [1,\infty) \right)

1083: \]

1084: then for some positive $c$ as $x\rightarrow \infty$ we have $

1085: \pr(X>x) \sim c x^{-\alpha}. $

1086: \end{fact}

1087: 

1088: Let $A'(a)$ denote the event that there is no cuttime so that the

1089: past has half-plane capacity between $1$ and $a$. Fact \ref{bert}

1090: implies

1091:  %

1092: \begin{equation}\label{alpha}

1093: \pr A'(a) \sim c a^{-\alpha}\quad\mbox{ as } a \rightarrow \infty.

1094: \end{equation}

1095:  %

1096: In order to conclude Theorem \ref{exponent}, we only need to show

1097: that half-plane capacity and time are not too far from each other;

1098: in fact it suffices to show the following.

1099: 

1100: \begin{lemma}

1101: We have $\pr(t/s<\capo\,B[0,t]<ts) \ge 1-ce^{-\gamma s}$.

1102: \end{lemma}

1103: 

1104: \begin{proof}

1105: For a set $A$ in the plane, let $M_x$ and $M_y$ denote the sup of

1106: the absolute value of the projection of $A$ to the $x$ and $y$

1107: axes, respectively. If $A\subset \overline \HH$ contains zero and

1108: is connected, then by (\ref{capim}) and considering the half-plane

1109: capacity of a rectangle we get

1110: $$

1111: M_y^2/4 \le \capo(A) \le c\max(M_y^2,M_x).

1112: $$

1113: Now let $A=B[0,1]$. Then it is easy to check the following simple

1114: property of the maxima of Brownian motion and the 3-dimensional

1115: Bessel processes.

1116: $$

1117: \pr( 1/s<M_y^2  \le \max(M_y^2,M_x)<s) \ge 1-ce^{-\gamma s},

1118: $$

1119: and the claim follows by scale-invariance.

1120: \end{proof}

1121: 

1122: \section{The value of the bead index $\alpha$}

1123: \lb{s.index}

1124: 

1125: Consider the process $B\sim \BE$; in this section we will index

1126: $B$ either by time or bead local time $\lambda$; in the latter

1127: case we will stick to the notation $\lambda$. Let $\mathcal L$

1128: denote Lebesgue measure, let $\mu$ denote the random measure on

1129: the half plane given by

1130: $$

1131: \mu(A)={\mathcal L}(\lambda\,:\,B(\lambda)\in A),

1132: $$

1133: and let $\overline \mu(A):=\ev \mu(A)$. By the scaling of

1134: half-plane capacity $a(\lambda)$ and the scale-invariance of

1135: Brownian motion we get that for $r\ge 0$

1136: $$

1137: \omu(rA) = r^{2\alpha} \omu(A).

1138: $$

1139: 

1140: 

1141: 

1142: \begin{lemma} \lb{strip}

1143: The $\omu$-measure of $\,\Strip_1=\RR \times (0,1]$

1144: is finite.

1145: \end{lemma}

1146: \begin{proof}

1147: Consider the random measure

1148: $$

1149: \mu(\Lambda,A)= {\mathcal L}(\lambda\in \Lambda\,:\,B(\lambda)\in

1150: A)

1151: $$

1152: and let $\omu(\Lambda,A):=\ev\mu(\Lambda,A)$. Let $\lambda$ be the

1153: first bead time at least $1$ so that $B(\lambda)\in \Strip_1$, and

1154: let $f$ denote the corresponding conformal shift. Let

1155: $p=\pr(\lambda<\infty)$; it is easy to check that $p<1$. For $n\ge

1156: 0$ we have

1157: \begin{eqnarray*}

1158: \ev(\mu([1,n+1],\Strip_1) \gi \GG_\lambda) &\le&

1159: \one(\lambda<\infty) \omu([0,n+1-\lambda],f(\Strip_1))\\&\le&

1160: \one(\lambda<\infty)\omu([0,n],\Strip_1)

1161: \end{eqnarray*}

1162: because of the cuttime Markov property (Proposition \ref {mp}) and

1163: the fact \re{shiftdown} that $f(\Strip_1)\subset \Strip_1$.

1164: Taking expected values gives

1165: $$\omu([1,n+1],\Strip_1) \le p

1166: \omu([0,n],\Strip_1)=:p \omu_n.$$ This yields the recursion

1167: $\omu_{n+1} \le 1 +p \omu_n$, which gives the bound $\omu_\infty

1168: \le 1/(1-p)$, as required.

1169: \end{proof}

1170: 

1171: \begin{lemma}\lb{absolute}

1172: $\omu$ is absolutely continuous with respect to Lebesgue measure

1173: on $\HH$ with density bounded below and above on compacts.

1174: \end{lemma}

1175: 

1176: \begin{proof}

1177: Let $A\subset \HH\setminus\{0\}$ be a closed set so that $A\cup

1178: \RR$ and $\HH \setminus A$ are connected, and let $f$ be the

1179: conformal homeomorphism $(\HH\setminus A,0,\infty)\to

1180: (\HH,0,\infty)$. Call such $f$ a subdomain map.

1181: 

1182: Fix a path $B$ which avoids $A$, and consider its image $f(B)$.

1183: The cuttimes $g$ of $B$ can be parameterized by $a(g)$ i.e.

1184: $\capo$ of the past, as well as $a'(g)$ that is $\capo$ of the

1185: past of the image $f(B)$. Fact \re{concap} implies that if

1186: $g_1<g_2$ are two cuttimes, then

1187: $$

1188: a(g_2)-a(g_1) \le a'(g_2)-a'(g_1),

1189: $$

1190: in particular, beads are smaller when measured by $a'$ then when

1191: measured by $a$. Let $E\subset \HH \setminus A$ a generic Borel

1192: subset, and let $\mu'(E)$ denote $\mu(E)$ measured for the

1193: $f$-mapping of the process $B$. Since a.s. $\mu(E)$ can be

1194: computed as the $\eps\to 0$ limit of the rescaled number of beads

1195: of size at least $\eps$ starting at a cutpoint in $E$, it follows

1196: that we have $\mu'(f(E))\le \mu(E)$. By the restriction property,

1197: $$

1198: \omu(f(E)) = \ev\left(\mu'(f(E))|B\mbox{ avoids }A\right) \le

1199: \ev(\mu(E)|B\mbox{ avoids }A),

1200: $$

1201: and therefore

1202: $$

1203: \omu(f(E)) \le \ev(\mu(E))/\pr(B\mbox{ avoids }A) =

1204: \omu(E)(f'(0)f'(\infty))^{-1}.

1205: $$

1206: Let $D_r(z)$ denote the open disk of radius $r$ about $z$. Let $K$

1207: be a compact subset of $\HH$. If $z\in K$ and $w$ is sufficiently

1208: close to $0$, then there exists a subdomain map $f$ so that

1209: $f(z)=w$. Consider the set of points $w$ for which $f$ exists for

1210: all $z\in K$; this set contains a rescaled version $sK$ of $K$. By

1211: compactness, we may choose subset maps for each $z\in K,sw\in sK$

1212: so that

1213: \begin{eqnarray*}

1214: c_0&\le& f'(0)f'(\infty),  \\

1215: D_{sr}(sw)&\subseteq &f(D_{c_2r}(z))\qquad\mbox{ for all }r<c_1

1216: \end{eqnarray*}

1217: with constants $c_0,c_1,c_2$ depending on $K$ only. Then

1218: \begin{eqnarray*}

1219: s^{2\alpha}\omu(D_{r}(w)) &=& \omu(D_{sr}(sw)) \\&\le&

1220: \omu(f(D_{c_2r}(z))) \\&\le&

1221: (f'(0)f'(\infty))^{-1}\omu(D_{c_2r}(z))\\&\le&

1222: c_0^{-1}\omu(D_{c_2r}(z)).

1223: \end{eqnarray*}

1224: This uniform bound implies the claim by standard arguments about

1225: absolute continuity.

1226: \end{proof}

1227: 

1228: The following are needed for the proof of

1229:  \newtheorem*{indexhalf.thm}{Theorem \ref{indexhalf}}

1230:  \begin{indexhalf.thm}

1231:  \indexhalfthm

1232:  \end{indexhalf.thm}

1233: 

1234: Let $D_r=D_r(i)$ denote the open disk of radius $r$ about $i$. Let

1235: $\tau$ be the (possibly infinite) first hitting time of $D_r$ for

1236: the process $B\sim\BE$. Let $B^{(1)}$ denote the path image

1237: $B[0,\tau]$. Recall the definition of the capacity $\capz(D,A)$ of

1238: $A$ from $\infty$ in $D$ from Section \ref{s.conformal}, and use

1239: the shorthand $\capz(t,A)= \capz(\rho B_t,A)$. Let

1240: $K_r=\capz(\tau,D_r)$, and recall the notation $x_r \asymp y_r$

1241: for the existence of a constant $c>0$ so that $c^{-1}y_r \le

1242: x_r\le c y_r$ (here for all small $r$).

1243: 

1244: \begin{proposition}\label{zbounds}

1245: As $r\to 0$ we have $ \omu(D_r) \asymp |\mbox{\rm log } r|\ev

1246: K_r^{1+2\alpha}. $

1247: \end{proposition}

1248: 

1249: Let $B^{(2)}$ denote the path after the last exit from $D_r$. Let

1250: $Z'_r=\pr (B^{(1)} \nis B^{(2)}\,|\,\FF_\tau)$, where ``$\nis$''

1251: denotes ``does not intersect''. Let $Z_r$ denote the probability

1252: given $\FF_\tau$ that the image of an independent process

1253: $\BE(\infty, D_r,\HH)$ does not intersect $B^{(1)}$; recall that

1254: this image can be defined via conformal mapping of $\BE$ started

1255: at a more conventional boundary point.

1256: 

1257: \begin{lemma}\lb{sizes} We have

1258: $|\mbox{\rm log}\ r| K_r \asymp Z_r \asymp Z'_r$, with

1259: deterministic constants.

1260: \end{lemma}

1261: 

1262: \begin{proof}

1263: We will use the time-reversal property of Brownian excursion: if

1264: $B\sim \BE(a,z,D)$ and ends at random time $\tau$, then

1265: $(B(\tau-t),\, t\in[0,\tau])\sim \BE(z,a,D)$. This, as well as

1266: conformal invariance, can be used to define the image of

1267: $\BE(\infty,z,\HH)$ (together with a time-parameterization

1268: starting at $-\infty$, but we won't need this).

1269: 

1270: The quantities in question are given by the measures of paths that

1271: do not intersect $B^{(1)}$ before hitting $D_r$ under the measures

1272: $|\mbox{log}\ r|\BE(\infty,\RR,\HH)$, $\BE(\infty, D_r,\HH)$, and

1273: $\BE(\infty,B(\tau),\HH)$, respectively. It follows from the

1274: definition of $\BE$ that the distribution of these paths up to the

1275: hitting time of $D_r$ given the hitting position agree. It is easy

1276: to check that the in each case hitting position has a smooth

1277: density bounded below and above with respect to uniform measure on

1278: $\partial D_r$. The $|\mbox{log}\ r|$ normalizing factor is

1279: necessary so that the measure of paths hitting $D_r$ is bounded

1280: below and above by constants as $r\to 0$.

1281: \end{proof}

1282: 

1283: \begin{proofof}{Theorem \ref{indexhalf}}

1284: By the Lemma \ref{absolute}, as $r\rightarrow 0$ we have

1285: $$

1286: \omu(D_r) \asymp r^2.

1287: $$

1288: We have

1289: $$

1290: \ev K_r^\gamma \asymp |\mbox{log }r|^{-\gamma} \ev Z_r^\gamma =

1291: r^{\xi(1,\gamma)+o(1)},$$ where the first approximation follows

1292: from Lemma \ref{sizes}, and the second is one definition of the

1293: intersection exponent. More precisely, we should consider the

1294: analogue of $Z_r$ for $B^{(1)'}$, which has distribution

1295: $\BE(0,D_r,\HH)$, but this is absolutely continuous with density

1296: bounded above and below with respect to the distribution of

1297: $B^{(1)}$ given that it hits $D_r$ (see the proof of Lemma

1298: \ref{sizes}).

1299: 

1300: Thus by Proposition \ref{zbounds} we get $

1301: r^{\xi(1,1+2\alpha)+o(1)} = r^2,$ and the theorem follows by the

1302: monotonicity of $\xi(1,\cdot)$ and the known value $\xi(1,2)=2$

1303: (see \cite{lawler95} for these properties of $\xi$).

1304: \end{proofof}

1305: 

1306: \begin{proofof}{Proposition \ref{zbounds}}\ \\

1307: \indent{\bf Upper bound.} We write

1308: $$

1309: \ev(\mu(D_r)\,|\,\FF_\tau) = \ev(\mu(D_r) \,|\, B^{(1)}\nis

1310: B^{(2)}, \FF_\tau) \pr (B^{(1)} \nis B^{(2)}\,|\,\FF_\tau).

1311: $$

1312: Let $f$ denote the conformal shift at the first cuttime $g$ after

1313: time $\tau$. We have

1314: $$K_r=\capz(\tau,D_r)\ge \capz(g,D_r)=\capz(f(D_r)).

1315: $$

1316: Here, by slight abuse of notation, $f(D_r)$ denotes the image of

1317: the part of $D_r$ on which $f$ is defined. By \re{capdia} the

1318: above implies that $f(D_r)\subset \RR\times (0,cK_r].$ By scaling

1319: and Lemma \ref{strip} we have $\omu(\RR\times

1320: (0,K_r/2])=cK_r^{2\alpha}.$ Thus by the cuttime Markov property

1321: (Proposition \ref{mp}) we get

1322: $$

1323: \ev(\mu(D_r) \,|\, B^{(1)}\nis B^{(2)}, \FF_\tau) =

1324: \ev(\omu(f(D_r))\,|\,\FF_\tau) \le cK_r^{2\alpha}.

1325: $$

1326: Here and in the sequel $c$ denotes a constant whose value may

1327: change by line to line. By Lemma \ref{sizes}

1328: $$

1329: \ev(\mu(D_r) \gi \FF_\tau) \le c Z'_rK_r^{2\alpha} \le c|\mbox{log

1330: }r| K_r^{1+2\alpha},

1331: $$

1332: which implies the upper bound in Proposition \ref{zbounds}.

1333: 

1334: \bigskip

1335: {\bf Lower bound.} Let $A_1$ denote the event that $B[0,\tau]$

1336: does not intersect the set $ \{w\in D_{2r}\,:\,\arg(w-i)\in

1337: (\pi/4,7\pi/4)\}.$ It is easy to check (see, for example

1338: \cite{lwssharp}) that for all $\gamma\in [1,3]$ and $r<1/2$ we

1339: have

1340:  \bel{za1}

1341:  \ev(Z_r^\gamma;\,A_1) \ge c\ev Z_r^\gamma.

1342:  \ee

1343: Consider the event

1344: $$

1345: A=\{B(g)\in D_{r},\, B[0,g]\nis D_{r/2},\, \capz(g,D_{r/32})\ge

1346: cK_r \}.

1347: $$

1348: This implies $B^{(1)}\nis B^{(2)}$, and it is easy to check that

1349: there is an absolute constant  $c_1$ so that

1350: $$

1351: \pr(A\,|\,B^{(1)}\nis B^{(2)},\FF_\tau) \ge c_1 \one_{A_1}.

1352: $$

1353: We have

1354: \begin{eqnarray*}

1355:  \ev( \mu(D_{r}) \gi \FF_\tau) &\ge&

1356:  \ev(\mu(D_{r});\, A \gi \FF_\tau)\one_{A_1} \\&\ge&

1357:  \ev(\mu(D_{r}) \gi A,\FF_\tau)

1358:  \pr(A \gi B^{(1)}\nis B^{(2)}, \FF_\tau)

1359:  \pr(B^{(1)}\nis B^{(2)}\,|\, \FF_\tau)\one_{A_1}.

1360: \end{eqnarray*}

1361: 

1362: By the cuttime Markov property (Proposition \ref{mp}) we get

1363: $$

1364:  \ev(\mu(D_{r}) \gi A,\FF_\tau)

1365: = \ev(\omu(f(D_{r})) \gi A,\FF_\tau).

1366: $$

1367: The event $A$ implies that the domain of $f$ contains $D_{r/2}$,

1368: and therefore by the K\"obe quarter theorem $f(D_{r/2})$ contains

1369: a ball $D'$ of radius $|f'(i)|r/8$ centered at $f(i)$. By the same

1370: theorem applied to $f^{-1}$, we get that $f^{-1}(D')$ contains

1371: $D_{r/32}$. By monotonicity and conformal invariance of harmonic

1372: measure, we get

1373: $$

1374: \mbox{diam}(D') \ge c\capz(D')=c\capz(g,f^{-1}(D')) \ge

1375: c\capz(g,D_{r/32}) \ge c' K_r,

1376: $$

1377: where the last inequality is an assumption in $A$. Since

1378: $f(\partial D_r)$ contains an curve $K$ that separates $0$ and

1379: $D'$ from $\infty$ in $\HH$, we have

1380: $$

1381: \dist(D',0)\le \mbox{diam}(K)\le c\,\capz(K)\le

1382: c\,\capz(f(D_r))\le cK_r.

1383: $$

1384: When $K_r=1$, this implies that $\omu(D')$ is bounded below by a

1385: constant. The scaling property then implies that in general

1386: $\omu(D')\ge cK_r^{2\alpha}$. Putting all the above together and

1387: using Lemma \ref{sizes} we get

1388: $$ \ev (\mu(D_{r})\,|\,\FF_\tau) \ge c

1389: Z'_r K_r^{2\alpha}\one_{A_1} \ge c  |\mbox{log }

1390: r|K_r^{1+2\alpha}\one_{A_1}

1391: $$ and the bound follows from (\ref{za1}).

1392: \end{proofof}

1393: \subsection*{Open questions and conjectures}

1394: 

1395: \begin{question}\rm

1396: The parameter $a_0$ in the conformal shift \re{shiftexp} defines

1397: the ``horizontal'' location of a cutpoint. It follows from our

1398: proofs that the process $((a(\lambda),a_0(\lambda)),\, \lambda \ge

1399: 0)$ is a 2-dimensional L\'evy process stable under scaling with

1400: exponents $1/2$ and $1$ in the two respective coordinates. In

1401: particular $(a_0(\lambda),\, \lambda>0)$ is a Cauchy process. What

1402: is the joint distribution of the two processes? (This question may

1403: be better considered in the $\mbox{SLE}_6$ framework.)

1404: \end{question}

1405: 

1406: \begin{question}\rm

1407: Is it possible to interpret bead local time as a local time at

1408: zero of some process? This may be better answered by considering a

1409: version of $\mbox{SLE}_6$.

1410: \end{question}

1411: 

1412: \begin{conjecture}\rm

1413: Consider the $\sigma$-finite bead process conditioned to survive

1414: up to distance $r$ (or time $r^2$, or imaginary part $r$), and let

1415: $r\rightarrow \infty$. The limiting process exists and has the

1416: restriction property. The exponent (Werner, personal

1417: communication) should equal 2 (see \cite{lwsrest} for

1418: definitions).

1419: \end{conjecture}

1420: 

1421: \begin{question}\rm

1422: The bead time $\lambda$ defines a random measure on the

1423: half-plane. The closed support of this measure is the set of

1424: cutpoints. Is it possible to derive the Hausdorff dimension of the

1425: set of cutpoints using this measure, or more generally, using

1426: beads? This would give a new, conceptual proof for the value of

1427: the intersection exponent $\xi(1,1)=5/4$.

1428: \end{question}

1429: 

1430: \bigskip

1431: \noindent {\bf Acknowledgments.} The author thanks Jim Pitman for

1432: stimulating remarks and recommending the reference \cite{gp80},

1433: which can also be used to construct bead local time. He also

1434: thanks Oded Schramm and Wendelin Werner for inspiring discussions

1435: and conscientious referee for several important remarks and

1436: corrections.

1437: \bibliography{bbb}

1438: 

1439: \end{document}

1440: