1: 

2: % LaTeX2e

3: % Planar aggregation and the coalescing Brownian flow

4: % James Norris & Amanda Turner

5: %

6: \documentclass[12pt]{article}

7: %	PACKAGES

8: \usepackage{amsmath}

9: \usepackage{amssymb}

10: \usepackage{amsthm}

11: \usepackage{latexsym}

12: \usepackage{color,fancyhdr,lastpage,graphicx}

13: \usepackage{subfigure, epsfig, epsf}

14: \usepackage{xspace}

15: \usepackage{setspace}

16: \usepackage{bbm}

17: 

18: \graphicspath{{thesis/Figures/}}

19: 

20: %	NUMBERING SYSTEM

21: \theoremstyle{plain}

22: \newtheorem{thm}{Theorem}[section]

23: \newtheorem{lemma}{Lemma}[section]

24: \newtheorem{defn}[lemma]{Definition}

25: \newtheorem{prop}[lemma]{Proposition}

26: \newtheorem{theorem}[lemma]{Theorem}

27: \newtheorem{proposition}[lemma]{Proposition}

28: \newtheorem{corollary}[lemma]{Corollary}

29: \theoremstyle{remark}

30: \newtheorem{rem}[lemma]{Remark}

31: \newtheorem{notn}[lemma]{Notation}

32: \newtheorem{example}[lemma]{Example}

33: 

34: %       MACROS

35: \def\bg{\begin{color}{blue}}

36: \def\br{\begin{color}{red}}

37: \def\eg{\end{color}}

38: \def\er{\end{color}}

39: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

40: % Put shorter definition first to include extra footnotes %%%

41: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

42: \def\note{\footnote{\bg\comment\eg}}

43: \def\note{}

44: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

45: \def\PP{\mathbb{P}}

46: \def\ve{{\varepsilon}}

47: \def\le{\leqslant}

48: \def\pd{\partial}

49: \def\ds{\pd_s}

50: \def\dt{\pd_t}

51: \def\du{\pd_u}

52: \def\dr{\pd_r}

53: \def\les{\preceq}

54: \def\ge{\geqslant}

55: \def\ges{\succeq}

56: \def\es{\emptyset}

57: \def\da{\downarrow}

58: \def\widebar{\overline}

59: \def\xto{\xrightarrow}

60: \def\E{{\mathbb E}}

61: \def\O{{\Omega}}

62: \def\Q{{\mathbb Q}}

63: \def\R{{\mathbb R}}

64: \def\C{{\mathbb C}}

65: \def\N{{\mathbb N}}

66: \def\T{{\mathbb T}}

67: \def\Z{{\mathbb Z}}

68: \def\S{{\Sigma}}

69: \def\a{{\alpha}}

70: \def\b{{\beta}}

71: \def\d{{\delta}}

72: \def\D{{\Delta}}

73: \def\t{{\tau}}

74: \def\k{{\kappa}}

75: \def\g{{\gamma}}

76: \def\G{{\Gamma}}

77: \def\s{{\sigma}}

78: \def\l{{\lambda}}

79: \def\L{{\Lambda}}

80: \def\th{{\theta}}

81: \def\o{{\omega}}

82: \def\z{{\zeta}}

83: \def\cA{{\cal A}}

84: \def\cS{{\cal S}}

85: \def\cL{{\cal L}}

86: \def\cR{{\cal R}}

87: \def\cC{{\cal C}}

88: \def\cE{{\cal E}}

89: \def\cF{{\cal F}}

90: \def\cH{{\cal H}}

91: \def\cV{{\cal V}}

92: \def\cD{{\cal D}}

93: \def\cZ{{\cal Z}}

94: \def\cW{{\cal W}}

95: \def\cM{{\cal M}}

96: \def\bx{{\text{\boldmath$x$}}}

97: \def\by{{\text{\boldmath$y$}}}

98: \def\bd{{\text{\boldmath$d$}}}

99: \def\bw{{\text{\boldmath$w$}}}

100: \def\bbX{{\text{\boldmath$\bar X$}}}

101: \def\bbY{{\text{\boldmath$\bar Y$}}}

102: \def\bbD{{\text{\boldmath$\bar D$}}}

103: \def\bbW{{\text{\boldmath$\bar W$}}}

104: \def\bW{{\text{\boldmath$W$}}}

105: \def\bD{{\text{\boldmath$D$}}}

106: \def\bX{{\text{\boldmath$X$}}}

107: \def\bY{{\text{\boldmath$Y$}}}

108: \def\tS{{\tilde S}}

109: \def\tx{{\tilde x}}

110: \def\tp{{\tilde p}}

111: \def\tq{{\tilde q}}

112: \def\tX{{\tilde X}}

113: \def\tL{{\tilde L}}

114: \def\tW{{\tilde W}}

115: \def\tD{{\tilde D}}

116: \def\half{{\frac12}}

117: \def\q{\quad}

118: \def\id{\operatorname{id}}

119: \def\cp{\operatorname{cap}}

120: \def\dom{\operatorname{dom}}

121: \def\supp{\operatorname{supp}}

122: \def\trace{\operatorname{trace}}

123: \def\Log{\operatorname{Log}}

124: \def\<{\langle}

125: \def\>{\rangle}

126: \def\ra{\rightarrow}

127: \def\ua{\uparrow}

128: \def\sse{\subseteq}

129: \def\sm{\setminus}

130: \def\fn{\footnote}

131: \renewcommand\epsilon{\ve}

132: %	MARGINS

133: \addtolength{\headheight}{0.4in}

134: \addtolength{\voffset}{-0.5in}

135: \addtolength{\textheight}{0.25in}

136: \addtolength{\footskip}{0.25in}

137: \addtolength{\evensidemargin}{-0.6in}

138: \addtolength{\oddsidemargin}{-0.6in}

139: \addtolength{\textwidth}{1.2in}

140: \addtolength{\headwidth}{1.2in}

141: 

142: 

143: \begin{document}

144: \bibliographystyle{plain}

145: 

146: %	FRONT MATTER

147: \begin{center}

148: \LARGE \textbf{Planar aggregation and the coalescing Brownian flow}

149: 

150: \vspace{0.2in}

151: 

152: \large {\bfseries James Norris

153: \footnote{Statistical Laboratory, Centre for Mathematical Sciences,

154:   Wilberforce Road, Cambridge, CB3 0WB, UK }

155: \& Amanda Turner

156: \footnote{Department of Mathematics and Statistics,

157: Fylde College,

158: Lancaster University,

159: Lancaster,

160: LA1 4YF, UK}}

161: 

162: \vspace{0.2in}

163: \small

164: \today

165: 

166: \end{center}

167: \begin{abstract}

168: We study a scaling limit associated to a model of planar aggregation.

169: The model is obtained by composing certain independent random conformal maps.

170: The evolution of harmonic measure on the boundary of the cluster is shown to

171: converge to the coalescing Brownian flow.

172: \end{abstract}

173: \vspace{0.2in}

174: 

175: %	BODY OF ARTICLE

176: %\ct

177: 

178: \section{Introduction}

179: A simplified model of aggregation in two dimensions may be formulated as follows.

180: Let $K_0$ be the closed disc of radius $1$, with centre at the origin, and

181: let $P_1$ be another closed disc, of diameter $\d>0$ say,

182: tangent to $K_0$ at a point chosen uniformly at random.

183: Set $D_0=(\C\cup\{\infty\})\sm K_0$ and $D_1=(\C\cup\{\infty\})\sm K_1$,

184: where $K_1=K_0\cup P_1$.

185: Write $F_1$ for the unique conformal isomorphism $D_0\to D_1$ such that

186: $F_1(\infty)=\infty$ and $F_1'(\infty)>0$.

187: Suppose then that $F_2,F_3,\dots$ are independent and identically distributed

188: copies of $F_1$, and define

189: $$

190: D_n=F_1\circ\dots\circ F_n(D_0).

191: $$

192: The complementary sets $K_n=\C\sm D_n$ form an increasing family and may be

193: considered as a model of aggregation, where, for each $n\ge0$, we add at time $n+1$

194: a new particle $P_{n+1}=K_{n+1}\sm K_n$.

195: Note that $P_{n+1}$ is the image under the conformal map $F_1\circ\dots\circ F_n$

196: of a disc distributed as $P_1$ and independent of $K_n$.

197: Since harmonic measure is invariant under conformal maps,

198: conditional on $K_n$, the random point at which

199: $P_{n+1}$ is attached to $K_n$ is distributed on the boundary of $K_n$

200: according to the normalized harmonic measure from infinity. In this respect, the

201: model is appropriate for the aggregation of diffusive particles `coming in from infinity'.

202: We note however that the added particle is distorted in a way depending

203: on the current cluster $K_n$.

204: 

205: In this paper, we identify a scaling limit for the restrictions to the boundary of $D_0$

206: of the maps $(F_m\circ\dots\circ F_n)^{-1}$, for

207: $m\le n$, as $m,n\to\infty$ and $\d\to0$ in a suitable way.

208: This can be thought of as identifying the time-evolution of the total harmonic measure

209: carried by the various `fingers' of the growing cluster.

210: The limit is not sensitive to the shape of the first added particle: although we

211: specified discs in the description above, the same sort of limit applies more generally.

212: The limit object is the flow of coalescing Brownian motions on the circle, also

213: known as the Brownian web.

214: 

215: A brief review of related work is given next, followed by some illustrations of typical

216: clusters, for certain cases of the model. From Section \ref{LFC} on, we confine our attention

217: to the restrictions of maps to the boundary of $D_0$, generalizing at the same 

218: time to a natural class of L\' evy random flows on the circle. We first show

219: a weak convergence result

220: for the time-evolutions of finitely many points. In Section \ref{CBF} we

221: describe a new canonical space for the coalescing Brownian

222: flow.

223: Then, in Section \ref{SKOR}, a larger flow-space, of Skorokhod type,

224: is introduced. This is a complete separable metric space

225: suitable for the formulation of weak convergence of stochastic flows

226: which are not necessarily

227: continuous in time or space.

228: The convergence of L\' evy flows to the coalescing Brownian flow,

229: is shown in Section \ref{MR}. Time reversal of flows is discussed in Section

230: \ref{TR}. Then, in Section \ref{model}, 

231: convergence results for the aggregation model 

232: are deduced as corollaries of the convergence of L\' evy flows.

233: Some technical details are left to an Appendix, where we discuss, in

234: particular, the relation between our formulation of the coalescing Brownian flow

235: and the Brownian webs of Fontes et al. 

236: 

237: \section{Review of related work}

238: 

239: \subsection{Coalescing Brownian motions}

240: It is straightforward to define a finite family of (standard) Brownian motions 

241: in one dimension, with 

242: given space-time starting points, independent until they collide, and

243: coalescing on collision.

244: The possibility to extend such a model to a space-time continuum of starting

245: points was shown by Arratia in 1979 in his PhD thesis \cite{A79}, where the

246: model was considered as a limit object for coalescing random walks on the integers.

247: Subsequent work has been carried out by many people

248: including Harris \cite{harris}, who was interested in general coalescing

249: stochastic flows, and Piterbarg \cite{Piterbarg}, who showed that

250: Arratia's flow arises as a weak limit of rescaled isotropic

251: stochastic flows. Further properties were developed by T\'{o}th and Werner \cite{TW} in 1998,

252: who used the flow to contruct an object which they called

253: the `continuous true self-repelling motion'. 

254: Tsirelson \cite{Tsirelson} studied Arratia's flow, formulated in terms of $L^2$-spaces, rather than pathwise,

255: as an example in the general theory of stochastic flows. 

256: Fontes, Isopi,

257: Newman and Ravishankar \cite{FINR,FN} introduced the name `Brownian web'

258: in 2004,

259: to describe a number of new formulations of Arratia's flow, and gave further

260: characterization and convergence results. 

261: The paper \cite{FINR}

262: characterizes the Brownian web as a

263: random element of a space of compact collections of

264: paths with specified starting points. 

265: 

266: In this paper we follow most closely the viewpoint of T\'{o}th and Werner but lay greater

267: stress on the almost sure flow-type properties, formulating

268: Arratia's flow as a random variable in

269: a certain complete separable metric space of continuous weak flows. 

270: We show that there is a

271: unique Borel probability measure on this space with respect to which all

272: the $n$-point motions are coalescing Brownian motions, and that this 

273: measure is invariant under time reversal.

274: Moreover, we show that any sequence of random such flows, all of 

275: whose $n$-point motions

276: converge to coalescing Brownian motions, converges to the coalescing Brownian 

277: flow.  

278: The exact correspondence between our work and that in \cite{FINR} is discussed in the Appendix. 

279: 

280: \subsection{Planar random growth}

281: \label{dlasec}

282: 

283: Our motivation for looking at the coalescing Brownian flow arises from a

284: surprising connection with planar random growth processes.

285: 

286: The simplest sorts of planar growth process to formulate take values in finite

287: subsets of $\Z^2$, starting from a singleton at the origin, 

288: and grow by the successive addition of sites adjacent to the present cluster, which are

289: chosen according to some distribution determined by the present cluster.

290: In 1961, Eden \cite{Eden} introduced one such process, where the added site is simply chosen

291: uniformly from all adjacent sites. This has been 

292: considered as a model for the growth of bacterial cells or tissue cultures 

293: of cells that are constrained from moving.

294: In 1981, Witten and Sander \cite{W+S} put forward another such process,

295: known as diffusion-limited aggregation or DLA. Here, particles perform 

296: random walks `starting from infinity' 

297: until they reach a site adjacent to the cluster. This site is then added to the cluster.

298: DLA is considered as a model for the formation of aggregates by deposition,

299: such as soot particles.

300: A family of further processes, indexed by a parameter $\eta\in[0,1]$ and

301: called dielectric-breakdown models,

302: were discussed by Niemeyer et al. \cite{NPW}.

303: In these processes, the random walk hitting probabilities on

304: sites adjacent to the cluster are interpreted as an electric field $E$;

305: the probabilities of attachment for new sites are then chosen proportional to $E^\eta$.

306: Thus the case $\eta=0$ is the Eden model and the case $\eta=1$ is DLA.

307: 

308: The primary interest in these and other related processes is in the 

309: asymptotic behaviour of large clusters. Computational investigations reveal

310: the emergence of complex structures, of fractal type, which sometimes appear a good match

311: for observed phenomena. However, such

312: investigations are highly sensitive to model variations and there are few notable

313: mathematical results, with the exception of Kesten's 1987 

314: growth estimate \cite{Kesten} for DLA. Moreover, the

315: simulations appear to show that clusters depend on the fine

316: lattice structure at large scales. This suggests that lattice-based processes may never

317: properly model continuum phenomena.

318: 

319: In 1998, Hastings and Levitov \cite{HL} formulated a family of continuum growth models

320: in terms of sequences of iterated conformal maps, indexed by a parameter $\a\in[0,2]$. 

321: As in the model described in the Introduction, 

322: they identify the cluster after the arrival of $n$ particles with a conformal map 

323: $\Phi_n=F_1\circ\dots\circ F_n$ on $D_0$, where, conditional on $\Phi_n$, the map $F_{n+1}$ corresponds

324: to a particle attached at a random point $z_{n+1}$ on the unit circle, but now of radius $\d_{n+1}=\d_0|\Phi_n'(z_{n+1})|^{-\a}$.

325: Hastings and Levitov argue, by comparing local growth rates, that their model can be related to lattice 

326: dielectric-breakdown by setting $\eta=\alpha-1$. Further exploration of this relation is discussed in the 

327: survey paper by Bazant and Crowdy \cite{B+C}.

328: 

329: Carleson and Makarov \cite{C+M}, in 2001, obtained a growth estimate for a deterministic analogue 

330: of the DLA model. 

331: In 2005, Rohde and Zinsmeister \cite{RZ} considered

332: the case $\alpha=0$ in the Hastings--Levitov family. 

333: They established a scaling limit, in the case where the particle size is small (but fixed),

334: as the number of particles tends to infinity, and showed that the limit sets were one dimensional. They

335: also gave estimates for the dimension of the limit sets in the case of general $\alpha$,

336: and discussed limits of deterministic variants.

337: 

338: We also consider the case $\alpha=0$ but in the limiting regime where the particle

339: diameter $\d$ becomes small and the number of particles is of the order $\d^{-3}$.

340: We then show that the resulting flow map, restricted to points on the unit circle, 

341: converges to the coalescing Brownian flow.

342: 

343: \section{Some illustrations}

344: \label{illus}

345: 

346: In the Introduction we described the construction of a sequence of iterated conformal maps that can be

347: interpreted as a model of the aggregation of diffusive particles `coming in from infinity'. 

348: We chose the basic particle $P_1$ to be a closed disc of diameter $\d>0$. However our analysis 

349: turns out to apply to a wide range of basic particle shapes, as we shall see in Section \ref{model}.

350: Recall that $D_0$ denotes the complement of the closed unit disc $K_0$ in $\C\cup\{\infty\}$ and 

351: $D_1$ denotes the complement of $K_1$ in $\C\cup\{\infty\}$, where $K_1=K_0\cup P_1$. Write

352: $F_1$ for the conformal isomorphism $D_0\to D_1$ fixing $\infty$, with $F_1'(\infty)>0$ and write

353: $G_1$ for the inverse isomorphism $D_1\to D_0$. Denote the upper half-plane by $H$.

354: 

355: We now discuss briefly the form of $G_1$ for two particular choices of $P_1$.

356: The first of these, corresponding to the slit $[1,1+\d]$, was used to generate some

357: realizations of the cluster for various values of $\d$, which are presented in Figure \ref{DLAfig}.

358: 

359: \begin{figure}[p]

360:   \vspace{-20pt}

361:     \subfigure[The cluster after a few arrivals with $\d=1$.]

362:       {\label{DLAfew}

363:       \epsfig{file=dlafew.eps,width=5.8cm}}

364:     \hfill

365:     \subfigure[The cluster after 100 arrivals with $\d=1$.]

366:       {\label{DLAmany}

367:       \epsfig{file=dlamany.eps,width=5.8cm}}

368:     \subfigure[The cluster after 800 arrivals with $\d=0.1$.]

369:       {\label{slit10}

370:       \epsfig{file=slit10.eps,width=5.8cm}}

371:     \hfill

372:     \subfigure[The cluster after 5000 arrivals with $\d=0.04$.]

373:       {\label{slit25}

374:       \epsfig{file=slit25.eps,width=5.8cm}}

375:     \subfigure[The cluster after 20000 arrivals with $\d=0.02$.]

376:       {\label{slit50}

377:       \epsfig{file=slit50.eps,width=5.8cm}}

378:     \hfill

379:     \subfigure[The stochastic flow $(X_{t0})_{t \in [0,1]}$ with

380:     $\d=0.02$.]

381:       {\label{flow50}

382:       \epsfig{file=flow50.eps,width=5.8cm}}

383:   \caption{\textsl{The slit model case of simplified Hastings-Levitov DLA}}

384:   \label{DLAfig}

385: \end{figure}

386: 

387: For the slit, we can obtain the map $G_1$ by composing the sequence of maps

388: $$

389: D_1\to H\sm (0,i\sqrt{t}]\to H\to H\to D_0,

390: $$

391: given by

392: $$

393: g_1(z)=i\frac{z-1}{z+1},\q

394: g_2(z)=\sqrt{z^2+t},\q

395: g_3(z)=\l z,\q

396: g_4(z)=\frac{i+z}{i-z}.

397: $$

398: Here, we choose $t=\d^2/(2+\d)^2$ so that $g_2\circ g_1(1+\d)=g_2(i\d/(2+\d))=0$ 

399: and choose $\l=1/\sqrt{1-t}$ so that

400: $g_3\circ g_2\circ g_1(\infty)=g_3\circ g_2(i)=\l i\sqrt{1-t}=i$. Then $G_1(1+\d)=1$

401: and $G_1(\infty)=\infty$, as required.

402: The maps extend to the boundaries of their domains in obvious ways. In particular the boundary

403: of the slit maps under $G_1$ to a boundary arc $[-\th_\d,\th_\d]$ of the unit circle. 

404: Note that $G_1(1)=g_4(\pm\l\sqrt{t})$ and $g_4'(0)=-2i$, from which we see that 

405: \begin{equation}\label{DTHN}

406: \frac{\th_\d}\d\to1\q\text{as}\q \d\to0. 

407: \end{equation}

408: 

409: We observe in Figure \ref{DLAmany}, when $\d=1$, that incoming particles are markedly distorted 

410: and, in particular, that particles arriving later tend to be larger. This effect is diminished

411: when we examine smaller values of $\d$. In Figure \ref{slit50}, the cluster is a rough ball, but with

412: some sort of internal structure. The colours label arrivals in different epochs, showing that growth

413: over time is uniform and there is a close relationship between time of arrival and the distance

414: from the origin at which the particle sticks. Figure \ref{flow50} focuses on the motion of points

415: on the original circular boundary, which is the aspect of these simulations examined theoretically

416: in this paper. The coalescing motion displayed agrees with our theoretical predictions. The size of

417: the gaps between the flow lines gives the amount of harmonic measure carried on fingers of the

418: cluster based between the chosen points on the unit circle.

419: 

420: \def\keep{  $$

421:   F_1^{-1}(z) = \left ( 1 - \left ( \frac{1}{1+2N}\right )^2 \right )

422:   \frac{z+1}{2z} \left ( z + 1 + \sqrt{z^2 + 1 - 2z \frac{ 1 +\left (

423:   \frac{1}{1+2N}\right )^2}{1 - \left ( \frac{1}{1+2N}\right )^2}

424:   }\right ) - 1.

425:   $$

426:   This map can be extended continuously to the boundary of the unit disc by setting $F_1^{-1}(e^{2 \pi i x}) = \exp (2 \pi f(x))$, for $x \in (0, 1)$, where

427:   $$

428:   f(x) =  \begin{cases}

429:           \pi^{-1} \tan^{-1} \sqrt{\frac{\tan^2 \pi x + (1 +

430:           2N)^{-2}}{1 - (1 + 2N)^{-2}}} & \quad x \in (0,

431:          \frac{1}{2}) \\

432:          - \pi^{-1} \tan^{-1} \sqrt{\frac{\tan^2 \pi x + (1 +

433:           2N)^{-2}}{1 - (1 + 2N)^{-2}}} & \quad x \in (\frac{1}{2},1).

434:          \end{cases}

435:   $$

436: }

437: 

438: In the case where $P_1$ is the lune $L=\{z\in\C:|z-1|\le\d,|z|>1\}$, the map $G_1$

439: is given by

440: $$

441: G_1(z)=\frac{(e^{i\theta}z-1)^a-(e^{-i\theta}z-1)^a}{(z-e^{-i\theta})^a-(z-e^{i\theta})^a},

442: $$

443: where

444: $$

445: \theta=2\tan^{-1}\d\sqrt{1-\d^2/4},

446: \q a=\frac{\pi}{\pi-\cos^{-1}(\d/2)}.

447: $$

448: We shall use in Section \ref{model} the following estimate of the 

449: logarithmic capacity of $K_1$ for this case. As $\d\to0$, 

450: \begin{equation}\label{LCE}

451: \cp(K_1)=-\log\lim_{z\rightarrow\infty}\frac{G_1(z)}z

452:                        =\log\frac{a\sin\theta}{\sin(a\theta)}=2\d^2+o(\d^2).

453: \end{equation}

454: 

455: \section{A class of L\' evy flows on the circle}\label{LFC}

456: We introduce a class of random flows on the circle, whose distributions

457: are invariant under rotations of the circle, and, under which, each point

458: on the circle performs a L\' evy process of mean $0$ with no diffusive part.

459: The flow maps are in general not continuous on the circle, but have a

460: non-crossing property. In a certain asymptotic regime, the

461: motion of the flow from a countable family of starting points

462: is shown to converge weakly to a family of

463: coalescing Brownian motions.

464: 

465: We specify a particular flow by the choice of a non-decreasing,

466: right-continuous function $f^+:\R\to\R$ with the following {\em degree}

467: $1$ property\footnote{These functions can be considered as liftings of maps from the circle

468: $\R/\Z$ to itself having an order-preserving property.

469: In practice, the circle map will be a perturbation of the identity map and

470: our basic map $f^+$ will be the unique lifting which is close to the identity map on $\R$.}

471: \begin{equation}

472: \label{circdef}

473: f^+(x+n) = f^+(x) + n, \quad x\in\R,\q n \in \mathbb{Z}.

474: \end{equation}

475: Denote the set of such functions by $\cR$ and write $\cL$ for the

476: analogous set of left-continuous functions.

477: Each $f^+\in\cR$ has a left-continuous modification $f^-\in\cL$, given by $f^-(x)=f(x-)$.

478: Write $\cD$ for the set of all pairs $f=\{f^-,f^+\}$.

479: When $f^+$ is continuous, we shall also write $f=f^+$ and, generally, we write $f$ in place of $f^\pm$

480: in expressions where the choice of left or right-continuous modification makes no difference to the value.

481: For $f^+\in\cR$ and $z=\tilde z+\Z\in\R/\Z$, define $(\t_z f)^+\in\cR$ by

482: $$

483: (\t_z f)^+(x)=\tilde z+f^+(x-\tilde z),\q x\in\R.

484: $$

485: The degree $1$ property ensures that there is no dependence on the choice of representative $\tilde z$.

486: The sets $\cR$ and $\cL$ are closed under composition, but $\cD$ is not.

487: In fact, if $f_1,f_2\in\cD$, then $f_2^-\circ f_1^-$ is the left-continuous modification of $f_2^+\circ f_1^+$

488: if and only if $f_1$ sends no interval of positive length to a point of discontinuity of $f_2$.

489: We say in this case that $f_2\circ f_1\in\cD$, denoting by $f_2\circ f_1$ the pair $\{f_2^-\circ f_1^-,f_2^+\circ f_1^+\}$.

490: Define $\id(x)=x$ and set $\cD^*=\cD\sm\{\id\}$.

491: We assume throughout that our basic map $f\in\cD^*$.

492: Write $\tilde f^\pm$ for the periodic functions $\tilde f^\pm(x)=f^\pm(x)-x$.

493: Define constants  $\rho=\rho(f)>0$ and $\b=\b(f)\in\R$

494: by

495: \begin{equation}

496: \label{lambdadef}

497: \rho \int_0^1\tilde{f}(x)^2 dx = 1, \q \beta =\rho\int_0^1\tilde{f}(x) dx.

498: \end{equation}

499: 

500: We now define a probability space $(\O,\cF,\PP)$. Take $\O$ to be the set

501: of those integer-valued Borel measures on $\R\times(\R/\Z)$

502: which are finite on bounded sets, and also have the property that

503: $\omega(\{t\} \times(\R/\Z)) \in \{0,1\}$ for all $t \in \mathbb{R}$.

504: We interpret the first factor $\R$ of the underlying space as time.

505: Write $\cF^o$ for the $\s$-algebra on $\O$ generated by evaluations on Borel sets.

506: For intervals $I\sse\R$, write $\cF_I^o$

507: for the $\s$-algebra on $\O$ generated by evaluations on Borel subsets of $I\times(\R/\Z)$.

508: There is a unique probability measure $\PP = \PP^{\rho}$ on

509: $(\O,\cF^o)$ which makes the coordinate map

510: $\mu(\o,dt,dz)=\o(dt,dz)$ into a Poisson random measure on

511: $\R\times(\R/\Z)$ with intensity $\nu(dt,dz)=\rho\, dtdz$.

512: Here $dz$ denotes Lebesgue measure on $\R/\Z$: we shall use,

513: without further comment, in defining integrals,

514: the obvious identification of $\R/\Z$ and $[0,1)$. 

515: Write $\cF$ for the completion of $\cF^o$ with respect to $\PP$,

516: extending $\PP$ to $\cF$ as usual.

517: 

518: We first construct the flow in the case where $\b=0$.

519: Given $\omega \in \Omega$, for each $t \in \mathbb{R}$ define $F_t=\{F_t^-,F_t^+\}\in \cD$ by

520: $$

521: F_t^\pm = \begin{cases}

522:           (\t_z f)^\pm, & \text{if $\o$ has an atom at $(t,z)$}, \\

523:           \id,          & \text{otherwise}.

524:          \end{cases}

525: $$

526: For each finite interval $I\sse\R$, define $X_I^+\in\cR$ and $X_I^-\in\cL$ by

527: $$

528: X_I^\pm=F_{T_n}^\pm\circ\dots\circ F_{T_1}^\pm,

529: $$

530: where $T_1<\dots<T_n$ are the times of the atoms of $\o$ in $I\times(\R/\Z)$.

531: We take $X_I^\pm=\id$ if there are no such atoms.

532: 

533: In the case where $\b\not=0$, we replace $\o$ in the preceding construction

534: by $\o^\b$, given by

535: $$

536: \o^\b(dt,dz)=\o(dt,d(z+\b t)),

537: $$

538: to obtain $X^{\b,\pm}_I$, and then set

539: $$

540: X_I^\pm(x)=X_I^{\b,\pm}(x+\b s)-\b t,

541: $$

542: where $s=\inf I$ and $t=\sup I$.

543: 

544: Since $f$ can have at most countably many points of discontinuity and intervals of

545: constancy, and since, under $\PP$, the positions of the atoms of $\o$ are

546: distributed uniformly on $\R/\Z$, we have, almost surely, $X_I=\{X_I^-,X_I^+\}\in\cD$ for all

547: intervals $I$.

548: Write $I=I_1\oplus I_2$ if $I_1, I_2$ are disjoint intervals with $\sup I_1=\inf I_2$ and $I=I_1\cup I_2$.

549: Note that $(X_I:I\sse\R)$ has the following properties: 

550: \begin{align}\label{LFM}

551: &X^+_I(x)\text{ and $X^-_I(x)$ are random variables for all finite intervals $I$ and all $x\in\R$},\\

552: \label{LWF}

553: &X_I^+=X_{I_2}^+\circ X_{I_1}^+\text{ and $X_I^-=X_{I_2}^-\circ X_{I_1}^-$ whenever $I=I_1\oplus I_2$},\\

554: \label{LPC}

555: &X^+_{(s,t)}(x)=X^-_{(s,t)}(x)=x-\b(t-s)\text{ for all $x\in\R$, eventually as $s\ua t$ or $t\da s$}.

556: \end{align}

557: 

558: Fix $e=(s,x)\in\R^2$, and define two processes $X^{e,-}_t$ and $X^{e,+}_t$,

559: both starting from $e$, by setting $X_t^{e,\pm}=X^\pm_{(s,t]}(x)$ for $t\ge s$.

560: Then $X^{e,-}$ and $X^{e,+}$ are both piecewise continuous, cadlag, and satisfy the integral equations

561: $$

562: X_t^{e,\pm}=x+\int_{(s,t]\times[0,1)}\tilde f^\pm(X^{e,\pm}_{r-}-z)(\mu-\nu)(dr,dz),

563: \q t\ge s.

564: $$

565: Under $\PP$, we have $X^{e,-}_t=X^{e,+}_t$ for all $t\ge s$, almost surely.

566: We shall therefore drop the $\pm$ from now on and write simply $X^e$.

567: Write $\mu^f_e$ for the distribution of $X^e$ under $\PP$

568: on the Skorokhod space $D_e=D_x([s,\infty),\R)$ of cadlag paths starting from $x$ at time $s$.

569: Write $\mu_e$ for the distribution on $D_e$ of a standard Brownian motion

570: starting from $e$.

571: 

572: In the context of planar aggregation $f$ describes the action, on the boundary of the unit disc, of the 

573: conformal map corresponding to the arrival of a particle. This action is descibed precisely in 

574: Section \ref{model}. Where particles are symmetric around the axis through their attachment point, $\b=0$. 

575: Asymmetric particles can give rise to non-zero values of $\b$ and in these cases a drift is induced for which we have compensated.

576: 

577: \begin{proposition}\label{LBM}

578: We have $\mu^f_e\to\mu_e$ weakly on $D_e$, uniformly in $f\in\cD^*$ as $\rho(f)\to\infty$.

579: \end{proposition}

580: \begin{proof}

581: We drop the superscript $e$ within the proof to lighten the notation.

582: Under $\PP$, the process $(X_t)_{t\ge s}$ is a martingale. In fact, it is a

583: L\' evy process, with characteristic exponent $\chi$ given by

584: $$

585: \chi(\th)=\rho\int_0^1\left\{e^{i\th\tilde f(z)}-1-i\th\tilde f(z)\right\}dz,\q\th\in\R.

586: $$

587: In particular, we have

588: $\E(|X_{t_1}-X_{t_2}|^2)=|t_1-t_2|$ for all $t_1,t_2\ge s$.

589: A standard criterion (see for example \cite[page 143]{B} or \cite[page 355]{MR1943877}) allows us

590: to deduce that the family of laws $(\mu^f_e:f\in\cD^*)$ is tight in $D_e$.

591: We note that the process $(X_t^2-t)_{t\ge s}$ is also a martingale,

592: and that the jumps of $(X_t)_{t\ge s}$ are bounded in absolute

593: value by $\|\tilde f\|=\sup_{x\in\R}|\tilde f(x)|$.

594: Let $\mu$ be any weak limit law for the limit $\rho\to\infty$.

595: Write $(Z_t)_{t\ge s}$ for the coordinate process on $D_e$.

596: Under $\mu$, by standard arguments,

597: both $(Z_t)_{t\ge s}$ and $(Z_t^2-t)_{t\ge s}$

598: are local martingales in the natural filtration of $(Z_t)_{t\ge s}$.

599: By using the fact that $f$ is non-decreasing, we can obtain for all $\rho\ge1$ the estimate

600: \begin{equation}\label{FR3}

601: \|\tilde f\|\le2\rho^{-1/3}.

602: \end{equation}

603: Hence $\mu$ is supported on continuous paths and must therefore be $\mu_e$

604: by L\' evy's characterization of Brownian motion.

605: \end{proof}

606: 

607: We next fix a sequence $E=(e_k:k\in\N)$ in $\R^2$, where $e_k=(s_k,x_k)$ say,

608: and consider the sequence of processes $X^E=(X^k:k\in\N)$, where $X^k=X^{e_k}$.

609: Then $X^E$ is a random variable in the complete separable

610: metric space $D_E=\prod_{k=1}^\infty D_{e_k}$, where we define the metric $d_E$ on $D_E$ by

611: $$

612: d_E(\xi,\xi')=\sum_{k=1}^\infty 2^{-k}\{d(\xi^k,{\xi'}^k)\wedge 1\},

613: $$

614: and where $d$ denotes appropriate instances of the Skorokhod metric.

615: Write $\mu^f_E$ for the distribution of $X^E$ on $D_E$ under $\PP$.

616: 

617: We consider now a limit in which the basic map $f$ is an increasingly well-localized perturbation of

618: the identity, where we quantify this property in terms of the smallest constant

619: $\l=\l(f)\in(0,1]$ such that

620: $$

621: \rho\int_0^1|\tilde f(x+a)\tilde f(x)|dx\le\l,\q a\in[\l,1-\l].

622: $$

623: Denote by $(Z^k_t)_{t\ge s_k}$ the $k$th coordinate process on $D_E$,

624: given by $Z_t^k(\xi)=\xi^k_t$, and consider the

625: filtration $(\cZ_t)_{t\in\R}$ on $D_E$, where $\cZ_t$ is

626: the $\s$-algebra generated by $(Z^k_s:s_k<s\le t\vee s_k,k\in\N)$.

627: Write $C_E$ for the (measurable) subset of $D_E$ where each coordinate path is continuous.

628: Define also

629: $$

630: T^{jk}=\inf\{t\ge s_j\vee s_k:Z_t^j-Z_t^k\in\Z\}.

631: $$

632: We are thinking of the paths $(Z^k_t)_{t\ge s_k}$ as liftings of paths in the circle. Thus

633: $T^{jk}$ is the collision time of these circle-valued paths.

634: The following is a convenient reformulation of a result of Arratia \cite{A79}.

635: \begin{proposition}\label{ARRA}

636: There exists a unique Borel probability measure $\mu_E$

637: on $D_E$ under which, for all $j,k\in\N$, the processes

638: $(Z^k_t)_{t\ge s_k}$ and $(Z^j_tZ^k_t-(t-T^{jk})^+)_{t\ge s_j\vee s_k}$ are

639: both continuous local martingales in the filtration $(\cZ_t)_{t\in\R}$.

640: \end{proposition}

641: Of course $\mu_E$ is supported on $C_E$ and may naturally be considered as a measure

642: defined there.

643: We sketch a proof. For existence, one can take independent

644: Brownian motions from each of the given time-space starting points

645: and then impose a rule of coagulation on collision, deleting the path of lower index.

646: The law of the resulting

647: process has the desired properties. On the other hand, given a probability

648: measure such as described in the proposition, on some larger

649: probability space, one can use a supply of independent Brownian motions

650: to resurrect the paths deleted at each collision. Then L\' evy's

651: characterization can be used to see that one has recovered

652: the set-up used for existence. This gives uniqueness.

653: 

654: \begin{proposition}\label{LAF}

655: We have $\mu^f_E\to\mu_E$ weakly on $D_E$,

656: uniformly in $f\in\cD^*$, as $\rho(f)\to\infty$ and $\l(f)\to0$.

657: \end{proposition}

658: \begin{proof}

659: The families of marginal laws $(\mu^f_{e_k}:f\in\cD^*)$ are all tight, as shown

660: in Proposition \ref{LBM}.

661: Hence the family of laws $(\mu^f_E:f\in\cD^*)$ is also tight.

662: Let $\mu$ be any weak

663: limit law for the limits $\rho\to\infty$ and $\l\to0$. Under $\PP$, for $j,k$

664: distinct, the process

665: $$

666: X^j_tX^k_t-\int_{s_j\vee s_k}^tb(X^j_s,X^k_s)ds, \q t\ge s_j\vee s_k,

667: $$

668: is a martingale, where

669: $$

670: b(x,x')=\rho\int_0^1\tilde f(x-z)\tilde f(x'-z)dz.

671: $$

672: We have $|b(x,x')|\le\l$ whenever $\l\le|x-x'|\le1-\l$. Hence, by standard

673: arguments, under $\mu$, the process $(Z^j_tZ^k_t:s_j\vee s_k\le t<T^{jk})$ is a local

674: martingale. We know from the proof of Proposition \ref{LBM} that, under $\mu$, the processes

675: $(Z^j_t:t\ge s_j)$, $((Z^j_t)^2-t:t\ge s_j)$ and $(Z^k_t:t\ge s_k)$ are continuous local martingales.

676: But $\mu$ inherits from the laws $\mu^f_E$ the property

677: that, almost surely, for all $n\in\Z$, the process $(Z^j_t-Z^k_t+n:t\ge s_j\vee s_k)$ does not change sign.

678: Hence, by an optional stopping argument, $Z^j_t-Z^k_t$ is constant for $t\ge T^{jk}$.

679: It follows that $(Z^j_tZ^k_t-(t-T^{jk})^+)_{t\ge s_j\vee s_k}$ is a continuous local martingale.

680: Hence $\mu=\mu_E$, by Proposition \ref{ARRA}.

681: \end{proof}

682: 

683: We note that the characterizing property of $\mu_E$ is invariant under a permutation of the

684: sequence $(e_k:k\in\N)$, as is the topology of the metric space $D_E$. 

685: Hence we can define for any countable set $E\sse\R^2$, a unique Borel probability measure 

686: on $\mu_E$ on $D_E=\prod_{e\in E}D_e$, having the given property. Then, by Proposition \ref{LAF}, the 

687: distribution $\mu_E^f$ of $X^E=(X^e:e\in E)$ on $D_E$ converges weakly to $\mu_E$ as $\rho(f)\to\infty$

688: and $\l(f)\to0$.

689: 

690: \section{A new state-space for the coalescing Brownian flow}

691: \label{CBF}

692: The preceding statement is unsatisfactory in that it expresses convergence

693: of our L\' evy flows only for a given countable set of time-space starting points. To remedy this, we

694: must first formulate a suitable limit object. 

695: This object is known as Arratia's flow, or the Brownian web,

696: and has been studied in some depth. However, we have found it convenient to introduce a new

697: state-space of flows, which we now describe.

698: 

699: We begin by defining a metric on $\cD$.

700: Let $\cS$ denote the set of all periodic contractions on $\mathbb{R}$ having period 1.

701: Each $f\in\cD$ can be identified with some $f^\times\in\cS$ by drawing new

702: axes at an angle $\pi/4$ with the old, and scaling

703: appropriately. See Figure \ref{Phimap}.

704: \begin{figure}[ht]

705:   \centering

706:   \epsfig{file=diagmapcross.eps, width=8cm}

707:   \caption{\textsl{The map $f^\times$ obtained from $f$ by rotating

708:   the axes by $\frac{\pi}{4}$.}}

709: \label{Phimap}

710: \end{figure}

711: More formally, since $x+f^+(x)$ is strictly increasing in $x$, there is for each $t\in\R$

712: a unique $x\in\R$ such that

713: $$

714: \frac{x+f^-(x)}2\le t\le\frac{x+f^+(x)}2.

715: $$

716: Define $f^\times(t)=t-x$. Note that $\id^\times=0$.

717: Then {\em the map $f\mapsto f^\times:\cD\to\cS$ is a bijection},

718: so we can define a metric $d_\cD$ on $\cD$ by

719: $$

720: d_\cD(f,g)=\|f^\times-g^\times\|=\sup_{t\in[0,1)}|f^\times(t)-g^\times(t)|.

721: $$

722: A proof of the italicized assertion is given in the Appendix. The same is true for some 

723: further technical assertions which will be made below, written also in italics. 

724: The metric space $(\cS,\|\dots\|)$ is complete and locally compact, 

725: so the same is true for $(\cD,d_\cD)$.

726: An alternative characterization\footnote{Thus, $d_\cD$ is a close relative

727: of the L\' evy metric sometimes used on the set of distribution functions for real random

728: variables. The relationships of such a metric to the operations of composition and inversion 

729: in $\cD$, which are significant for us, do not appear to have been studied.}

730: of the metric $d_\cD$ is as follows: {\em for $f,g\in\cD$ and $\ve>0$, we have}

731: $$

732: d_\cD(f,g)\le\ve\iff f^-(x-\ve)-\ve\le g^-(x)\le g^+(x) \le f^+(x+\ve)+\ve

733: \text{ {\em for all }}x\in\R.

734: $$

735: We deduce that, for $f,g\in\cD$,

736: $$

737: d_\cD(f,g)\le\|f-g\|,\q 2d_\cD(f,\id)=\|f-\id\|,

738: $$

739: and

740: $$

741: d_\cD(f,g\circ f)\le\|g-\id\|\text{ when $g\circ f\in\cD$},\q

742: d_\cD(f,f\circ g)\le\|g-\id\|\text{ when $f\circ g\in\cD$}.

743: $$

744: Moreover, {\em for any sequence $(f_n:n\in\N)$ in $\cD$,

745: $$

746: f_n\to f\iff f_n(x)\to f(x)\text{ at every point $x$ where $f$ is continuous}.

747: $$}

748: Here and below, we write $f_n\to f$ to mean convergence in the metric $d_\cD$.

749: 

750: We now define our space of flows. We call them weak flows to emphasise that the

751: usual flow property may fail at points of spatial discontinuity.

752: Consider $\phi=(\phi_{ts}:s,t\in\R,s<t)$, with $\phi_{ts}\in\cD$ for all $s,t$.

753: Say that $\phi$ is a {\em weak flow} if

754: $$

755: \phi_{ut}^-\circ\phi_{ts}^-\le\phi_{us}^-\le\phi_{us}^+\le\phi_{ut}^+\circ\phi_{ts}^+,

756: \q s<t<u.

757: $$

758: Say that $\phi$ is {\em continuous} if, for all $t\in\R$,

759: $$

760: \phi_{ts}\to\id\text{ as $s\ua t$},\q\phi_{ut}\to\id\text{ as $u\da t$}.

761: $$

762: Write $C^\circ(\R,\cD)$ for the set of all continuous weak flows\footnote{\label{CDZ}In Section \ref{SLAM} we shall work with

763: a modified space of flows $C^\circ((0,\infty),\cD_0)$. Here we restrict to intervals $I\sse(0,\infty)$, and we take

764: $\phi_I\in\cD_0$, where $\cD_0$ is the set of circle maps whose liftings are in $\cD$. Given the continuity of the map (\ref{PHIC})

765: and the requirement $\phi_{ss}=\id$, there is an obvious identification of this space with $C^\circ((0,\infty),\cD)$}.

766: It will be convenient sometimes to extend a continuous weak flow $\phi$ to the diagonal, which we do

767: by setting $\phi_{ss}=\id$ for all $s\in\R$. 

768: Then, {\em for any $\phi\in C^\circ(\R,\cD)$, the map

769: \begin{equation}\label{PHIC}

770: (s,t)\mapsto\phi_{ts}:\{(s,t):s\le t\}\to\cD

771: \end{equation}

772: is continuous.}

773: 

774: Define, for $\phi,\psi\in C^\circ(\R,\cD)$,

775: $$

776: d_C(\phi,\psi)=\sum_{n=1}^\infty2^{-n}\{d_C^{(n)}(\phi,\psi)\wedge1\},

777: $$

778: where

779: $$

780: d_C^{(n)}(\phi,\psi)=\sup_{s,t\in(-n,n),s<t}d_{\cD}(\phi_{ts},\psi_{ts}).

781: $$

782: Then $d_C$ is a metric on $C^\circ(\R,\cD)$, {\em under which this space is

783: complete and separable}.

784: {\em For this metric, for all $s,t\in\R$ with $s<t$, and all $x\in\R$, the {\rm evaluation map}

785: $$

786: \phi\mapsto\phi_{ts}^+(x):C^\circ(\R,\cD)\to\R

787: $$ 

788: is continuous.

789: Moreover the Borel $\s$-algebra on $C^\circ(\R,\cD)$ is generated by

790: the set of all such evaluation maps with $s,t$ and $x$ rational.}

791:                                                                                                                                                                             

792: {\em For $e=(s,x)\in\R^2$ and $\phi\in C^\circ(\R,\cD)$, the map

793: $$

794: t\mapsto\phi_{ts}^+(x):[s,\infty)\to\R

795: $$

796: is continuous.} Hence we can define a measurable map

797: $Z^{e,+}:C^\circ(\R,\cD)\to C_e=C_x([s,\infty),\R)$ by

798: $$

799: Z^{e,+}(\phi)=(\phi_{ts}^+(x):t\ge s).

800: $$

801: Given a countable set $E\sse\R^2$, we then define a measurable map 

802: $Z^{E,+}:C^\circ(\R,\cD)\to C_E=\prod_{e\in E}C_e$ by

803: $$

804: Z^{E,+}(\phi)^e=Z^{e,+}(\phi).

805: $$

806: Set $C_E^{\circ,+}=\{Z^{E,+}(\phi):\phi\in C^\circ(\R,\cD)\}$.

807: Define similarly $Z^{e,-}$, $Z^{E,-}$ and $C_E^{\circ,-}$ and set $C_E^\circ=C_E^{\circ,+}\cap C_E^{\circ,-}$.

808: 

809: The following result translates into the language of continuous weak flows a result of T\'oth and Werner \cite[Theorem 2.1]{TW}, 

810: which itself was a variant of a result of Arratia \cite{A79}.

811: We shall give a complete proof, in part because we need most components of the proof also for our main convergence result,

812: and in part because our framework leads to some simplifications, for example in the probabilistic underpinnings

813: contained in Proposition \ref{MPI}.

814: The formulation in terms of continuous weak flows has advantages, in 

815: leading to a unique object, with a natural time-reversal invariance, and for the derivation of weak limits.

816: Recall that, for a countable set $F\sse\R^2$, we denote by $\mu_F$ the law on $C_F$ of a family of coalescing Brownian motions starting

817: from $F$, as discussed at the end of the preceding section.

818: 

819: \begin{theorem}\label{UBPR}

820: There exists a unique Borel probability measure $\mu_A$ on $C^\circ(\R,\cD)$

821: such that, for any finite set $F\sse\R^2$, we have

822: \begin{equation}\label{MUW}

823: \mu_A\circ (Z^{F,+})^{-1}=\mu_F.

824: \end{equation}

825: Moreover, for all $e\in\R^2$, we have, $\mu_A$-almost surely, $Z^{e,+}=Z^{e,-}$. 

826: \end{theorem}

827: \begin{proof}

828: Fix a countable subset $E$ of $\R^2$ containing $\Q^2$.

829: Consider $\phi,\psi\in C^\circ(\R,\cD)$ and suppose that $Z^{E,+}(\phi)=Z^{E,+}(\psi)$. Then $\phi_{ts}^+(x)=\psi_{ts}^+(x)$ for all

830: $s,x\in\Q$ and all $t>s$. This extends to all $x\in\R$ because $\phi_{ts}^+$ and $\psi_{ts}^+$ are both right continuous,

831: and then to all $s\in\R$ using (\ref{PHIC}), so $\phi=\psi$. 

832: Hence the map $Z^{E,+}:C^\circ(\R,\cD)\to C_E^{\circ,+}$ is a bijection.

833: Write $\Phi^{E,+}$ for the inverse bijection $C^{\circ,+}_E\to C^\circ(\R,\cD)$. 

834: Similarly, write $\Phi^{E,-}$ for the inverse of the bijection $Z^{E,-}:C^\circ(\R,\cD)\to C_E^{\circ,-}$.

835: Then {\em $C^{\circ,+}_E$ and $\Phi^{E,+}$ are

836: measurable, we have $\mu_E(C^\circ_E)=1$, and $\Phi^{E,+}=\Phi^{E,-}$ on $C^\circ_E$.} 

837: So we can define a Borel probability measure $\mu_A$ on $C^\circ(\R,\cD)$ by

838: $$

839: \mu_A=\mu_E\circ(\Phi^{E,+})^{-1}.

840: $$

841: Then

842: $$

843: \mu_E=\mu_A\circ(Z^{E,+})^{-1}.

844: $$

845: For $F\sse E$, we have $Z^{F,+}=\pi_{E,F}\circ Z^{E,+}$,

846: and $\mu_F=\mu_E\circ\pi_{E,F}^{-1}$,

847: where $\pi_{E,F}:C_E\to C_F$ is the obvious projection.

848: So

849: \begin{equation*}

850: \mu_F=\mu_A\circ(Z^{E,+})^{-1}\circ\pi_{E,F}^{-1}=\mu_A\circ(Z^{F,+})^{-1}.

851: \end{equation*}

852: On the other hand, for $E\sse E'$, we have $Z^E=\pi_{E',E}\circ Z^{E'}$,

853: and $\mu_E=\mu_{E'}\circ\pi_{E',E}^{-1}$.

854: Then $\pi_{E',E}$ restricts to a bijection 

855: $C^{\circ,+}_{E'}\to C^{\circ,+}_E$, so we obtain that

856: $$

857: \mu_A=\mu_{E'}\circ\pi_{E',E}^{-1}\circ\Phi_E^{-1}=\mu_{E'}\circ\Phi_{E'}^{-1}.

858: $$

859: This shows that $\mu_A$ does not depend on $E$, and so (\ref{MUW})

860: holds for all finite sets $F$.

861: This property then characterizes $\mu_A$ by a standard $\pi$-system argument.

862: Finally, $\mu_A$-almost surely, $Z^{E,+}\in C_E^\circ$, so $Z^{E,+}=Z^{E,-}$,

863: so $Z^{e,+}=Z^{e,-}$ for all $e\in E$.

864: \end{proof}

865: 

866: We call any $C^\circ(\R,\cD)$-valued random variable with law $\mu_A$ a {\em coalescing Brownian flow}.

867: The relationship of this space and measure to the Brownian web of Fontes et al. is explored in the Appendix.

868: 

869: \section{A Skorokhod-type space of non-decreasing flows on the circle}

870: \label{SKOR}

871: Since our L\'evy flows are not continuous in time, it will be necessary to

872: introduce a larger flow space to accommodate them.

873: Consider now $\phi=(\phi_I:I\sse\R)$, where $\phi_I\in\cD$ and $I$ ranges

874: over all non-empty finite intervals. 

875: Recall that we write $I=I_1\oplus I_2$ if $I_1, I_2$ are disjoint intervals with $\sup I_1=\inf I_2$ and $I=I_1\cup I_2$. 

876: Say that $\phi$ is a {\em weak flow} if,

877: \begin{equation}\label{WF}

878: \phi_{I_2}^-\circ\phi_{I_1}^-\le\phi_I^-\le\phi_I^+\le\phi_{I_2}^+\circ\phi_{I_1}^+,

879: \q I=I_1\oplus I_2.

880: \end{equation}

881: Say that $\phi$ is {\em cadlag}\footnote{This definition is more symmetrical in time than

882: is usual for `cadlag': a more accurate acronym would be {\em laglad}}

883: if, for all $t\in\R$,

884: $$

885: \phi_{(s,t)}\to\id\q\text{as $s\uparrow t$},\q

886: \phi_{(t,u)}\to\id\q\text{as $u\downarrow t$}.

887: $$

888: Write $D^\circ(\R,\cD)$ for the set of cadlag weak flows.

889: It will be convenient to extend a cadlag weak flow $\phi$ to the empty interval by setting $\phi_\es=\id$.

890: Given a finite interval $I$ and a sequence of finite intervals $(I_n:n\in\N)$, write $I_n\to I$ if

891: $$

892: I=\bigcup_n\bigcap_{m\ge n}I_m=\bigcap_n\bigcup_{m\ge n}I_m.

893: $$

894: {\em For any $\phi\in D^\circ(\R,\cD)$, we have}

895: \begin{equation}\label{PHID}

896: \phi_{I_n}\to\phi_I\q\text{{\em as} $I_n\to I$}.

897: \end{equation}

898: 

899: Let $\phi$ be a cadlag weak flow and suppose that $\phi_{\{t\}}=\id$ for all $t\in\R$.

900: Then, using \eqref{WF}, we have $\phi_{(s,t)}=\phi_{(s,t]}=\phi_{[s,t)}=\phi_{[s,t]}$ for all $s<t$ and, denoting

901: all these functions by $\phi_{ts}$, the family $(\phi_{ts}:s,t\in\R,s<t)$ is a continuous weak

902: flow in the sense of the preceding section.

903: 

904: For $\phi,\psi\in D^\circ(\R,\cD)$ and $n\ge1$, define

905: $$

906: d_D^{(n)}(\phi, \psi) = \inf_\lambda\left\{

907: \gamma(\lambda) \vee \sup_{I\sse\R}\|\chi_n(I)\phi_I^\times-\chi_n(\l(I))\psi_{\lambda(I)}^\times\|\right\},

908: $$

909: where the infimum is taken over the set of increasing homeomorphisms

910: $\lambda$ of $\R$, where

911: $$

912: \gamma(\lambda) = \sup_{t\in\R}|\l(t)-t|\vee\sup_{s,t\in\R,s<t}\, \left|\log\left(\frac{\lambda(t) - \lambda(s)}{t-s}\right)\right|,

913: $$

914: and where $\chi_n$ is the cutoff function\footnote{As in the case of the standard Skorokhod topology, localization

915: in time sits awkwardly with the stretching of time introduced via the homeomorphisms $\l$. There is no fundamental

916: obstacle, just some messiness at the edges. We note that, when $I\cup\l(I)\sse[-n,n]$, we have

917: $$

918: \|\chi_n(I)\phi_I^\times-\chi_n(\l(I))\psi_{\lambda(I)}^\times\|=d_\cD(\phi_I,\psi_{\l(I)}).

919: $$

920: Also, for all intervals $I$, we have $|\chi_n(\l(I))-\chi_n(I)|\le\g(\l)$ and

921: $$

922: \|\chi_n(I)\phi_I^\times-\chi_n(\l(I))\psi_{\lambda(I)}^\times\|

923: \le\chi_n(I)d_\cD(\phi_I,\psi_{\l(I)})+|\chi_n(\l(I))-\chi_n(I)|\|\psi^\times_{\l(I)}\|.

924: $$}

925: given by

926: $$

927: \chi_n(I)=0\vee(n+1-R)\wedge1,\q R=\sup I\vee(-\inf I).

928: $$

929: Then define

930: \begin{equation}\label{SMET}

931: d_D(\phi, \psi)=\sum_{n=1}^\infty2^{-n}\{d_D^{(n)}(\phi,\psi)\wedge1\}.

932: \end{equation}

933: Then $d_D$ is a metric on $D^\circ(\R,\cD)$ {\em under which $D^\circ(\R,\cD)$ is

934: complete and separable.} 

935: Moreover {\em the metrics $d_C$ and $d_D$ generate the same topology on $C^\circ(\R,\cD)$}.

936: {\em For the metric $d_D$, for all finite intervals $I$ and all $x\in\R$, the {\rm evaluation map}

937: $$

938: \phi\mapsto\phi_I^+(x):D^\circ(\R,\cD)\to\R

939: $$

940: is Borel measurable.

941: Moreover the Borel $\s$-algebra on $D^\circ(\R,\cD)$ is generated by

942: the set of all such evaluation maps with $I=(s,t]$ and $s,t$ and $x$ rational.}

943:                                                                                                                                                                             

944: {\em For $e=(s,x)\in\R^2$ and $\phi\in D^\circ(\R,\cD)$, the map

945: $$

946: t\mapsto\phi_{(s,t]}^+(x):[s,\infty)\to\R

947: $$

948: is cadlag.} Hence we can extend the map $Z^{e,+}$, which we defined on $C^\circ(\R,\cD)$ in the

949: preceding section, to a measurable map

950: $Z^{e,+}:D^\circ(\R,\cD)\to D_e=D_x([s,\infty),\R)$ by setting

951: $$

952: Z^{e,+}(\phi)=(\phi_{(s,t]}^+(x):t\ge s).

953: $$

954: Given a countable set $E\sse\R^2$, we then define a measurable map

955: $Z^{E,+}:D^\circ(\R,\cD)\to D_E$ by

956: $$

957: Z^{E,+}(\phi)^e=Z^{e,+}(\phi),

958: $$

959: and set 

960: $$

961: D_E^{\circ,+}=\{Z^{E,+}(\phi):\phi\in D^\circ(\R,\cD)\}.

962: $$

963: Define similarly $Z^{e,-}$, $Z^{E,-}$ and $D_E^{\circ,-}$, and set $D_E^\circ=D_E^{\circ,+}\cap D_E^{\circ,-}$.

964: 

965: \section{Convergence of L\' evy flows to the coalescing Brownian flow}

966: \label{MR}

967: We defined in Section \ref{LFC}, for a given basic map $f\in\cD^*$, on the canonical probability

968: space $(\O,\cF,\PP)$ of a Poisson random measure, a certain

969: random flow $(X_I:I\sse\R)$, with $X_I\in\cD$ for all finite intervals $I$.

970: The properties (\ref{LFM}), (\ref{LWF}) and (\ref{LPC}), noted above, imply that

971: $X(\o)\in D^\circ(\R,\cD)$ for $\PP$-almost all $\o\in\O$, and that the function

972: $X:\O\to D^\circ(\R,\cD)$ is measurable for the Borel $\s$-algebra of $D^\circ(\R,\cD)$.

973: Moreover, we know that

974: \begin{equation}\label{PME}

975: Z^{e,+}(X)=Z^{e,-}(X)\q\text{almost surely, for all $e\in\R^2$}.

976: \end{equation}

977: We denote by $\mu^f_A$ the law of $X$ on the Borel $\s$-algebra of $D^\circ(\R,\cD)$.

978: 

979: \def\comment{It may be possible to base an alternative proof, using a more direct tightness

980: argument on $D^\circ(\R,\cD)$, on the following estimates, which have remained unused. First

981: $$

982: \E(X_t^4)=3t^2\rho\int_0^1\tilde f(z)^2dz+t\rho\int_0^1\tilde f(x)^4dz\le3t^2+4t\rho^{-2/3},

983: $$

984: Then

985: $$

986: |X_{t0}(x)-x|\le\max_{k=0,1,\dots,n-1}|X_{t0}(k/n)-k/n|+1/n,

987: $$

988: so 

989: $$

990: 2\|d_\cD(X_{t0},\id)\|_{L^p(\PP)}\le(n\E(X_t^p))^{1/p}+1/n.

991: $$

992: Then, optimizing over $n$, we get

993: $$

994: \|d_\cD(X_{t0},\id)\|_{L^p(\PP)}\le2\E(X_t^p)^{1/(p+1)},

995: $$

996: provided the RHS is no greater than $1$. In particular

997: $$

998: \|d_\cD(X_{t0},\id)\|_{L^4(\PP)}\le2(3t^2+4t\rho^{-2/3})^{-1/5}.

999: $$}

1000: \begin{theorem}\label{MAIN}

1001: We have $\mu^f_A\to\mu_A$ weakly on $D^\circ(\R,\cD)$,

1002: uniformly in $f\in\cD^*$, as $\rho(f)\to\infty$ and $\l(f)\to0$.

1003: \end{theorem}

1004: \begin{proof}

1005: Take $E=\Q^2$.

1006: Consider $\phi,\psi\in D^\circ(\R,\cD)$ and suppose that $Z^{E,+}(\phi)=Z^{E,+}(\psi)$. Thus $\phi_{(s,t]}^+(x)=\psi_{(s,t]}^+(x)$ for all

1007: $s,x\in\Q$ and all $t>s$. This extends to all $x\in\R$ because $\phi_{(s,t]}^+$ and $\psi_{(s,t]}^+$ are both right continuous. Then,

1008: using (\ref{PHID}), we obtain $\phi_I(x)=\psi_I(x)$ for all finite intervals $I$.

1009: Hence the map $Z^{E,+}:D^\circ(\R,\cD)\to D^{\circ,+}_E$ is a bijection.

1010: Write $\Phi^{E,+}$ for the inverse bijection $D^{\circ,+}_E\to D^\circ(\R,\cD)$. 

1011: Similarly, write $\Phi^{E,-}$ for the inverse of the bijection $Z^{E,-}:D^\circ(\R,\cD)\to D^{\circ,-}_E$.

1012: Then {\em $D^{\circ,+}_E$ and $\Phi^{E,+}$ are

1013: measurable, and we have $\Phi^{E,+}=\Phi^{E,-}$ on $D^\circ_E$}.

1014: Write $\Phi^E$ for the common restriction of these functions to $D_E^\circ$.

1015: {\em Then $\Phi^E$ is continuous at $z$ for all $z\in C_E^\circ$.}

1016: We see from (\ref{PME}) that $\mu_E^f(D^\circ_E)=1$ for all $f\in\cD^*$, 

1017: and we recall from the proof of Theorem \ref{UBPR} that $\mu_E(C^\circ_E)=1$.

1018: 

1019: Consider the limit where $\rho(f)\to\infty$ and $\l(f)\to0$.

1020: By Proposition \ref{LAF}, we have $\mu^f_E\to\mu_E$ weakly on $D_E$.

1021: Then, by a standard weak convergence argument, see for example \cite{B}, we obtain 

1022: $\mu^f_A=\mu^f_E\circ(\Phi^E)^{-1}\to\mu_E\circ(\Phi^E)^{-1}=\mu_A$ weakly on $D^\circ(\R,\cD)$.

1023: \end{proof}

1024: 

1025: \section{Time reversal}\label{TR}

1026: For $f^+\in\cR$ and $f^-\in\cL$, we define a {\em left-continuous inverse}

1027: $(f^+)^{-1}\in\cL$ and a {\em right-continuous inverse} $(f^-)^{-1}\in\cR$

1028: by

1029: $$

1030: (f^+)^{-1}(y)=\inf\{x\in\R:f^+(x)>y\},\q (f^-)^{-1}(y)=\sup\{x\in\R:f^-(x)<y\}.

1031: $$

1032: The map $f^+\mapsto(f^+)^{-1}:\cR\to\cL$ is a bijection, with

1033: $((f^+)^{-1})^{-1}=f^+$ and

1034: $$

1035: (f_1^+\circ f_2^+)^{-1}=(f_2^+)^{-1}\circ(f_1^+)^{-1},\q f_1,f_2\in\cR.

1036: $$

1037: We have $f^+\circ (f^+)^{-1}=\id$\, if and only if $f^+$ is a homeomorphism.

1038: Define for $f=\{f^-,f^+\}\in\cD$ the {\em inverse}

1039: $f^{-1}=\{(f^+)^{-1},(f^-)^{-1}\}\in\cD$. Note that $(f^{-1})^\times=-f^\times$,

1040: so the map $f\mapsto f^{-1}:\cD\to\cD$ is an isometry.

1041: 

1042: Define the {\em time-reversal map} $\phi\mapsto\hat\phi$ on $D^\circ(\R,\cD)$ by

1043: $$

1044: \hat\phi_I=\phi_{-I}^{-1},

1045: $$

1046: where $-I=\{-x:x\in I\}$.

1047: It is straightforward to check that this is a well-defined isometry of

1048: $D^\circ(\R,\cD)$, which restricts to an isometry of $C^\circ(\R,\cD)$.

1049: Write $\hat\mu_A^f$ for the law of $\widehat{X^f}$, where

1050: $X^f\sim\mu_A^f$ is the L\'evy flow

1051: with basic map $f\in\cD^*$.  

1052: \begin{proposition}

1053: We have $\hat\mu_A^f=\mu_A^{f^{-1}}$ for all $f\in\cD^*$.

1054: \end{proposition}

1055: \begin{proof}

1056: Fix $f\in\cD^*$. Set $g=f^{-1}$ and

1057: $$

1058: \D=\{(x,y)\in\R^2:y<f(x)\}=\{(x,y)\in\R^2:x>g(y)\},\q \D_0=\{(x,y)\in\R^2:y<x\}.

1059: $$

1060: Then, by Fubini's theorem,

1061: $$

1062: \int_0^1\tilde f(x)dx=\int_0^1\int_\R(1_\D-1_{\D_0})(x,y)dxdy=-\int_0^1\tilde g(y)dy

1063: $$

1064: and

1065: $$

1066: \int_0^1\tilde f(x)^2dx=\int_0^1\int_\R2(y-x)(1_\D-1_{\D_0})(x,y)dxdy=\int_0^1\tilde g(y)^2dy.

1067: $$

1068: So $\rho(g)=\rho(f)$ and $\b(g)=-\b(f)$.

1069: Given $\o\in\O$, define $\hat\o\in\O$ by $\hat\o(I\times B)=\o((-I)\times B)$,

1070: for intervals $I\sse\R$ and Borel sets $B\sse(\R/\Z)$. The map $\o\mapsto\hat\o$

1071: is a measure preserving transformation of $(\O,\cF,\PP)$. We have also

1072: $(\hat\o)^\b=\widehat{\o^{-\b}}$. Now, suppressing the $\pm$ notation, for any finite interval

1073: $I$,

1074: $$

1075: X^f_{-I}(\hat\o,x)=\hat F_{\hat T_n}\circ\dots\circ\hat F_{\hat T_n}(x-\b t)+\b s,

1076: $$

1077: where $\hat T_1<\dots<\hat T_n$ are the times of the atoms of $(\hat\o)^\b$

1078: in $-I$, and where

1079: $$

1080: \hat F_t = \begin{cases}

1081:           \t_z f, & \text{if $(\hat\o)^\b$ has an atom at $(t,z)$}, \\

1082:           \id,          & \text{otherwise}.

1083:          \end{cases}

1084: $$

1085: Set

1086: $$

1087: G_t(y) = \begin{cases}

1088:           \t_zg, & \text{if $\o^{-\b}$ has an atom at $(t,z)$}, \\

1089:           \id,          & \text{otherwise}.

1090:          \end{cases}

1091: $$

1092: Note that $\hat T_k=-T_{n-k+1}$, where $T_1<\dots<T_n$ are

1093: the times of the atoms of $\o^{-\b}$ in $I$,

1094: and $(\t_zf)^{-1}=\t_zg$, so $(\hat F_{\hat T_k})^{-1}=G_{T_{n-k+1}}$.

1095: Then

1096: $$

1097: \widehat{X^f_I}(\hat\o,y)=(X^f_{-I})^{-1}(\hat\o,y)

1098: =G_{T_n}\circ\dots\circ G_{T_1}(y-\b s)+\b t=X^g_I(\o,y).

1099: $$

1100: Hence $X^g$ and $\widehat{X^f}$ have the same distribution.

1101: \end{proof}

1102: 

1103: We get as a corollary the reversibility of the limit, which is already known

1104: in various guises. See, for example, \cite{A79},\cite{FN} and \cite{TW}.

1105: 

1106: \def\comment{

1107: I have been puzzling over the question whether $\l(f)\to0$ implies $\l(f^{-1})\to0$ in general.

1108: This would tie things up neatly. As it stands, using reversibility, one of these

1109: conditions will suffice, with $\rho(f)\to\infty$, to give $\mu^f_A\to\mu_A$.

1110: I am pretty much convinced that $\l(f)\not=\l(f^{-1})$ in general but have not actually done

1111: an example to establish this.}

1112: 

1113: \begin{corollary}

1114: The law $\mu_A$ of the coalescing Brownian flow is invariant under time-reversal.

1115: \end{corollary}

1116: \begin{proof}

1117: Write $\hat\mu_A$ for the image measure of $\mu_A$ under time reversal.

1118: Fix $r>0$ and define $f\in\cD$ by

1119: $$

1120: f^+(n+x)=n+(x\vee r),\q n\in\Z,\q x\in[0,1).

1121: $$

1122: Then $\tilde f^+(x)=(r-x)\vee 0$ for $x\in[0,1)$, so $\rho(f)=3/r^3$ and

1123: $$

1124: \int_0^1\tilde f(x)\tilde f(x+a)dx=0,\q r\le a\le1-r,

1125: $$

1126: so $\l(f)\le r$. Moreover $\rho(f^{-1})=\rho(f)$ and $\l(f^{-1})=\l(f)$\note. 

1127: Consider the limit $r\to0$. By Theorem \ref{MAIN},

1128: we know that $\mu^f_A\to\mu_A$ and $\mu^{f^{-1}}_A\to\mu_A$, weakly on $D^\circ(\R,\cD)$.

1129: Since the time-reversal map $\phi\mapsto\hat\phi$ is an isometry, it

1130: follows, using the preceding proposition, that $\mu^{f^{-1}}_A=\hat\mu^f_A\to\hat\mu_A$,

1131: weakly on $D^\circ(\R,\cD)$. Hence $\mu_A=\hat\mu_A$.

1132: \end{proof}

1133: 

1134: 

1135: 

1136: \section{Scaling limit of the aggregation model}\label{SLAM}

1137: \label{model}

1138: 

1139: We now return to the aggregation model described in the Introduction and deduce from Theorem \ref{LAF}

1140: that, as the particle diameter $\d\to0$ and time is suitably speeded up, then the evolution of 

1141: harmonic measure on the fingers of the cluster converges to the coalescing Brownian flow.

1142: In fact, this scaling limit applies more generally and depends on the shape of the attached particles 

1143: through only a single scalar parameter, which determines the speed of the associated flow.

1144: We therefore begin this section by giving a more general formulation of the aggregation model than that

1145: in the Introduction.

1146: 

1147: Let $K_0$ denote the closed unit ball in $\C$ with centre at $0$. Set $D_0=(\C\cup\{\infty\})\sm K_0$.

1148: Let $P$ be a closed, connected, simply connected subset of $\overline{D_0}$ of diameter $\d\in(0,1]$ 

1149: such that $P\cap K_0=\{1\}$. Set $D=(\C\cup\{\infty\})\sm(K_0\cup P)$. 

1150: The set $P$ models an incoming particle, which is attached to $K_0$ at $1$. 

1151: Write $F$ for the unique conformal isomorphism $F:D_0\to D$ such that $F(\infty)=\infty$

1152: and $F'(\infty)>0$. Let $R_1,R_2,\dots$ be a sequence of random rotations of the plane about the origin

1153: and set $F_n=R_n\circ F\circ R_n^{-1}$. Now construct the aggregation model from the sequence $F_1,F_2,\dots$

1154: as in the Introduction.

1155: 

1156: Write $G$ for the inverse isomorphism $D\to D_0$.

1157: There exists a unique $g\in\cD$ such that $g$ restricts to a continuous map from the interval $(0,1)$ to itself, and such that

1158: $$

1159: G(e^{2\pi ix})=e^{2\pi ig(x)},\q x\in(0,1).

1160: $$

1161: Define $\rho(P)=\rho(g)>0$ and $\b(P)=\b(g)\in\R$ as at (\ref{lambdadef}). 

1162: Note that, if $P$ is symmetric in the real axis, then $g$ is an odd function, so $\b(P)=0$.

1163: 

1164: 

1165: Set $\G_n=G_n\circ\dots\circ G_1$, where $G_n=F_n^{-1}$, so that $\G_n:D_n\to D_0$.

1166: The restriction of $\G_n$ to the boundary $\partial K_n=\partial D_n$, gives a natural parametrization

1167: of the boundary of the $n$th cluster by the unit circle. It has the property that, for $\xi,\eta\in\partial K_n$,

1168: the normalized harmonic measure $h$ (from $\infty$) of the positively oriented boundary segment from $\xi$ to $\eta$

1169: is given by $\G_n(\eta)/\G_n(\xi)=e^{2\pi ih}$. 

1170: For $m,n\in\N$ with $m<n$, set 

1171: $$

1172: \G_{nm}=G_n\circ\dots\circ G_{m+1}|_{\partial K_0}. 

1173: $$

1174: Set $\G_{nn}=\id$. The circle maps $\G_{nm}$ belong to $\cD_0$ (see Footnote \ref{CDZ})

1175: and have the flow property

1176: $$

1177: \G_{nm}\circ \G_{mk}=\G_{nk},\q k\le m\le n.

1178: $$

1179: The map $\G_{nm}$ expresses how the harmonic measure on $\partial K_m$ is transformed by the arrival of new particles

1180: up to time $n$. The following result identifies the asymptotic behaviour of this family of maps, for symmetric particles $P$,

1181: in the limit as the

1182: particle diameter $\d$ becomes small.\footnote{If $P$ is not symmetric, we obtain the same result once the definition of

1183: $\G_I$ is modified to 

1184: $$

1185: \G_I(e^{2\pi ix})=e^{-2\pi i\b t}\G_{nm}(e^{2\pi i(x+\b s)}),

1186: $$

1187: where $s=\inf I$ and $t=\sup I$.

1188: The cluster exhibits in this case a first order spinning at speed $\b=\b(P)$, with second order fluctuations given by the

1189: coalescing Brownian flow.}

1190: We embed in continuous time by defining, for an interval $I\sse(0,\infty)$,

1191: $\G_I=\G_{nm}$ where $m$ and $n$ are the smallest and largest integers, respectively, in $\rho(P)I$.

1192: Then $(\G_I:I\sse(0,\infty))$ is a random variable in $D^\circ((0,\infty),\cD_0)$. We denote its law by $\mu^P_A$.

1193: Write here $\mu_A$ for the law of the coalescing Brownian flow on $C^\circ((0,\infty),\cD_0)$.

1194: 

1195: \begin{theorem}

1196: We have $\mu_A^P\to\mu_A$ weakly on $D^\circ((0,\infty),\cD_0)$ as $\d(P)\to0$.

1197: \end{theorem}

1198: \begin{proof}

1199: By a standard argument, it will suffice to prove that $\tilde\mu_A^P\to\mu_A$ weakly on $D^\circ((0,\infty),\cD_0)$ as $\d(P)\to0$,

1200: where we obtain $\tilde\mu^P_A$ by a Poissonized embedding, taking $\tilde\G_{(s,t]}=\G_{N_tN_s}$, with $N$ a Poisson process

1201: of rate $\rho(P)$. In the light of Proposition \ref{GDL} below, the theorem is then a straightforward translation

1202: of Theorem \ref{MAIN}.

1203: \end{proof}

1204: 

1205: \begin{corollary}

1206: Let $x_1,\dots,x_n$ be a positively oriented set of points in $\R/\Z$ and set $x_0=x_n$.

1207: Set $K_t=K_{\lfloor\rho(P)t\rfloor}$.

1208: For $k=1,\dots,n$, write $H_t^k$ for the harmonic measure in $K_t$ of the boundary segment of all

1209: fingers in $K_t$ attached between $x_{k-1}$ and $x_k$.

1210: Let $(B_t^1,\dots,B_t^n)_{t\ge0}$ be a family

1211: of coalescing Brownian motions in $\R/\Z$ starting from $(x_1,\dots,x_n)$. 

1212: Then, in the limit $\d(P)\to0$, $(H_t^1,\dots,H_t^n)_{t\ge0}$ converges weakly in $D([0,\infty),[0,1]^n)$

1213: to $(B_t^1-B_t^0,\dots,B^n_t-B_t^{n-1})_{t\ge0}$.

1214: \end{corollary}

1215: 

1216: \begin{proposition}\label{GDL}

1217: There is a universal constant $C<\infty$ such that $\d^{-3}/C \leq \rho(P) \leq C\d^{-3}$  

1218: and $\l(P)\le C\d^{1/4}$.

1219: \end{proposition}

1220: \begin{proof}

1221: It is shown in Lawler \cite{L} that there exists some universal constant $c<\infty$ such that if 

1222: $c\d\leq x\leq1-c\d$, then

1223: $$

1224: |\tilde{g}(x)| \leq \frac{3 \cp(P \cup K_0)}{2\pi|1 - e^{2 \pi ix}|} = \frac{3\cp(P \cup K_0)}{4\pi \sin(\pi x)},

1225: $$

1226: where $\cp(P \cup K_0) = - \log G'(\infty)$ is the logarithmic capacity of $P \cup K_0$. If $K_1 \sse K_2$, 

1227: then $\cp(K_1) \leq \cp(K_2)$ (see, for example, \cite{L}), 

1228: and hence $\cp(P \cup K_0) \le\cp(L\cup K_0) \leq c'\d^2$ 

1229: for some constant $c' > 0$, by the estimate \eqref{LCE}.

1230: 

1231: Now suppose $x \in (-c\d, c\d)$. Then $g(-c\d) \leq g(x) \leq g(c\d)$ and hence, for some $c''>0$,

1232: $$

1233: |\tilde{g}(x)|

1234: \le|\tilde{g}(c\d)|\vee|\tilde{g}(-c\d)|+2c\d 

1235: \le\frac{3c'\d^2}{4\pi\sin(\pi c\d)}+2c\d

1236: \le c''\d.

1237: $$

1238: Hence, we have

1239: \begin{align*}

1240: \rho(g)^{-1}&=\int_0^1|\tilde{g}(x)|^2 dx

1241: \le2c\d(c''\d)^2+\int_{c\d}^{1-c\d}\frac{9c'^2\d^4}{16\pi^2\sin^2(\pi x)}dx\\

1242: &=2cc''^2\d^3+\frac{9c'^2\d^4}{8\pi^3}\cot(c\pi\d)=O(\d^3).

1243: \end{align*}

1244: 

1245: To show the other side of the inequality, recall the estimate \eqref{FR3}, which implies

1246: that $\rho(g)\le8\|\tilde{g}\|^{-3}$. So it is enough to show that there exists

1247: some $x \in [0,1)$ with $|\tilde{g}(x)| \geq c'''\d$ for some $c'''>0$.

1248: Define

1249: $$

1250: \tau_A = \inf \{t>0: B_t \in A \},

1251: $$

1252: where $B_t$ is a Brownian motion starting from infinity. Since Brownian motion is invariant under conformal transformations,

1253: $$

1254: g^+(0)-g^-(0)

1255: %=g^+(x^-)-g^-(x^+-1)

1256: =\PP(\t_P\leq\t_{K_0}).

1257: $$

1258: Since $P$ has diameter $\d$, there exist two points, $w, z \in \overline{P}$, 

1259: with $w \in \overline{P} \cap K_0$ and $|w-z| \geq \d/2$. Since $P$ is connected there is a path, 

1260: $l \sse \overline{P}$ joining $w$ and $z$. Let $r$ be the line joining 0 to $z$. Reflect $l$ in $r$ and call the reflection $l'$ and the reflection of point $w$, $w'$. Now

1261: $$

1262: 2 \mathbb{P}(\tau_l \leq \tau_{K_0})\geq \mathbb{P}(\tau_{l \cup l'} \leq \tau_{K_0} )

1263: $$

1264: and 

1265: $$

1266: \mathbb{P}(\tau_{l \cup l'} \leq \tau_{K_0} ) \geq \mathbb{P}(\tau_r \leq \tau_{K_0}) \geq 8c'''(|z|-1),

1267: $$

1268: by the limit \eqref{DTHN}, but also

1269: $$

1270: \mathbb{P}(\tau_{l \cup l'} \leq \tau_{K_0}) \geq \mathbb{P}(B_{\tau_{K_0}} \in \text{arc } (w-w')) \geq 8c'''|w-w'|

1271: $$

1272: for some constant $c'''>0$. Either $(|z|-1)$ or $|w-w'| \geq \d/4$ which finishes the argument.

1273: 

1274: Finally, to show that $\l(g)\le C\d^{1/4}$ observe that for $\d$ sufficiently small, 

1275: if $a \in [\d^{1/4}, 1- \d^{1/4}]$, then at most one of $x$ and $x+a$ lie in 

1276: $\cup_{n \in \mathbb{Z}}[n-c\d, n+c\d]$. Hence

1277: $$

1278: \rho\int_0^1|\tilde{g}(x+a)\tilde{g}(x)|dx

1279: \le2\rho c''\d\int_{c\d}^{1-c\d}\frac{3c'\d^2}{4\pi\sin(\pi x)}dx 

1280: =\frac{3\rho c'c''\d^3}{\pi^2}\log\tan(\pi c\d)

1281: =O(\d).

1282: $$

1283: Hence there exists some $C>0$ such that, for all $a \in [C\d^{1/4}, 1- C\d^{1/4}]$, we have

1284: $$

1285: \rho \int_0^1|\tilde{g}(x+a)\tilde{g}(x)|dx < C\d^{1/4}.

1286: $$

1287: \end{proof}

1288: 

1289: \section{Appendix}

1290: \subsection{Some properties of the space $\cD$ of non-decreasing functions

1291: of degree $1$}

1292: We give proofs in this subsection of a number of assertions made in Section \ref{CBF}.

1293: \begin{proposition}

1294: \label{fcrossbijection}

1295: The map $f\mapsto f^\times:\cD\to\cS$ is a well-defined bijection, with inverse given by

1296: $$

1297: f^-(x)=\inf\{t+f^\times(t):t\in\R,x=t-f^\times(t)\},\q

1298: f^+(x)=\sup\{t+f^\times(t):t\in\R,x=t-f^\times(t)\}.

1299: $$

1300: \end{proposition}

1301: \begin{proof}

1302: Recall that $f^\times(t)=t-x$, where $x$ is the unique point such that $f^-(x)\le2t-x\le f^+(x)$.

1303: The periodicity of $f^\times$ is an easy consequence of the degree $1$ condition. We now show that

1304: $f^\times$ is a contraction. Fix $s,t\in\R$ and suppose that

1305: $f^\times(s)=s-y$. Switching the roles of $s$ and $t$ if necessary, we may assume without

1306: loss that $x\ge y$. If $x=y$, then $f^\times(s)-f^\times(t)=s-t$. On the other hand, if $x>y$, then

1307: $2s-y\le f^+(y)\le f^-(x)\le2t-x$, so

1308: $$

1309: -(t-s)\le-(t-s)+(2t-x)-(2s-y)=f^\times(t)-f^\times(s)=(t-s)-(x-y)<t-s.

1310: $$

1311: In both cases, we see that $|f^\times(t)-f^\times(s)|\le|t-s|$. Hence $f^\times\in\cS$.

1312: 

1313: Suppose now that $g\in\cS$. Consider, for each $x\in\R$, the set

1314: $$

1315: I_x=\{t+g(t):t\in\R,x=t-g(t)\}.

1316: $$ 

1317: Since $g$ is a contraction, these sets are all intervals,

1318: and, since $g$ is bounded, they cover $\R$.

1319: For $x,y\in\R$ with $x>y$, and for $s,t\in\R$ with $x=t-g(t),y=s-g(s)$, we have

1320: $t-s-(g(t)-g(s))=x-y>0$, so $s\le t$, and so

1321: $$

1322: t+g(t)-(s+g(s))=t-s+(g(t)-g(s))\ge0.

1323: $$

1324: Define $h^+(y)=\sup I_y$ and $h^-(x)=\inf I_x$.

1325: We have shown that $h^+(y)\le h^-(x)$.

1326: Moreover, since the intervals $I_x$ cover $\R$, the functions $h^\pm$ must be

1327: the left-continuous and right-continuous versions of a non-decreasing function $h$,

1328: which then has the degree $1$ property, because $g$ is periodic.

1329: Thus $h\in\cD$.

1330: 

1331: For each $t\in\R$, we have $h^\times(t)=t-x$,

1332: where $2t-x\in I_x$, and so $2t-x=s+g(s)$ for some $s\in\R$

1333: with $x=s+g(s)$. Then $s=t$ and so $h^\times(t)=g(t)$. Hence $h^\times=g$.

1334: On the other hand, if we take $g=f^\times$ and if $x$ is a point of continuity of $f$,

1335: then we find $I_x=\{f(x)\}$, so $h^+(x)=h^-(x)=f(x)$. Hence $h=f$.

1336: We have now shown that $f\mapsto f^\times:\cD\to\cS$ is a bijection, and that its inverse

1337: has the claimed form.

1338: \end{proof}

1339: 

1340: \begin{proposition}

1341: For $f,g\in\cD$ and $\ve>0$,

1342: $$

1343: d_\cD(f,g)\le\ve\iff f^-(x-\ve)-\ve\le g^-(x)\le g^+(x) \le f^+(x+\ve)+\ve

1344: \text{ for all }x\in\R.

1345: $$

1346: Moreover, for any sequence $(f_n:n\in\N)$ in $\cD$,

1347: $$

1348: f_n\to f\text{ in }\cD\q\iff\q f_n^+(x)\to f(x)\text{ at all points $x\in\R$ where $f$ is continuous}.

1349: $$

1350: \end{proposition}

1351: \begin{proof}

1352: Suppose that $d_\cD(f,g)\le\ve$ and that $x$ is a continuity point of $g$.

1353: Then $g(x)=t+g^\times(t)$ for some $t\in\R$ with $x=t-g^\times(t)$.

1354: We must have $x+\ve\ge t-f^\times(t)$ and $g(x)\le t+f^\times(t)+\ve$,

1355: so $f^+(x+\ve)+\ve\ge t+f^\times(t)+\ve\ge g(x)$.

1356: Similarly $f^-(x-\ve)-\ve\le g(x)$. These inequalities extend to all $x\in\R$

1357: by taking left and right limits along continuity points.

1358: 

1359: Conversely, suppose that $t \in \R$ is such that $|f^\times(t)-g^\times(t)| = d_\cD(f,g)$

1360: and let $x = t-g^\times(t)$ and $y=t-f^\times(t)$. Then $x$ is the unique point with

1361: $g^-(x)+x \leq 2t \leq g^+(x)+x$ and $y$ is the unique point such that

1362: $f^-(y)+y \leq 2t \leq f^+(y)+y$. Hence $f^-(x-\ve)-\ve\le g^-(x)\le g^+(x) \le f^+(x+\ve)+\ve$

1363: implies $y \in [x - \ve, x+ \ve]$ and so $d_\cD(f,g) = |y-x| \le \ve$.

1364: 

1365: It follows directly that for any sequence $(f_n:n\in\N)$ in $\cD$,

1366: if $d_\cD(f_n,f) \rightarrow 0$ as $n \rightarrow \infty$,

1367: then $f_n^+(x)\to f(x)$ at all points $x\in\R$ where $f$ is continuous.

1368: 

1369: Now suppose $f_n^+(x)\to f(x)$ at all points $x\in\R$ where $f$ is continuous.

1370: By equicontinuity, it will suffice to show that $f_n^\times(t) \rightarrow f^\times(t)$ for each $t\in\R$.

1371: Set $x=t-f^\times(t)$ and $x_n=t-f^\times_n(t)$.

1372: Given $\ve>0$, choose $y_1\in(x-\ve,x)$ and $y_2\in(x,x+\ve)$, both points

1373: of continuity of $f$. Now $f(y_1)+y_1<2t<f(y_2)+y_2$, so there exists $N\in\N$

1374: such that for all $n \geq N$, we have $f^+_n(y_1)+y_1<2t<f^+_n(y_2)+y_2$, which implies

1375: $x_n\in[y_1,y_2]$ and hence $|f_n^\times(t) - f^\times(t)| < \ve$, as required.

1376: \end{proof}

1377: \begin{proposition}\label{WFL}

1378: Suppose $f_n\to f, g_n\to g, h_n\to h$ in $\cD$ with

1379: $h_n^+\le f_n^+\circ g_n^+$ for all $n$. Then $h^+\le f^+\circ g^+$.

1380: \end{proposition}

1381: \begin{proof}

1382: It will suffice to establish the inequality at all

1383: points $x$ where $g$ and $h$ are both continuous.

1384: Given $\ve>0$, since $f^+$ is right-continuous, there exists a point $y>g(x)$

1385: where $f$ is continuous and such that $f(y)<f^+(g(x))+\ve$.

1386: Then $f_n^+(y) < f^+(g(x))+\ve$ and $g_n^+(x)\le y$ eventually, so

1387: $$

1388: h_n^+(x)\le f_n^+(g_n^+(x))\le f_n^+(y)<f^+(g(x))+\ve

1389: $$

1390: eventually. Hence $h^+(x)=\lim_{n\to\infty}h_n^+(x)\le f^+(g^+(x))$,

1391: as required.

1392: \end{proof}

1393: 

1394: \subsection{Some properties of the the continuous flow-space $C^\circ(\R,\cD)$

1395: and cadlag flow-space $D^\circ(\R,\cD)$}

1396: We give proofs in this subsection of a number of assertions made in Sections \ref{CBF} and \ref{SKOR}.

1397: \begin{proposition}

1398: For $(s,x)\in\R^2$ and $\phi\in D^\circ(\R,\cD)$, the map

1399: $$

1400: t\mapsto\phi_{(s,t]}^+(x):[s,\infty)\to\R

1401: $$

1402: is cadlag, and is moreover continuous whenever $\phi\in C^\circ(\R,\cD)$.

1403: \end{proposition}

1404: \begin{proof}

1405: Given $t\ge s$ and $\ve>0$, we can choose $\d>0$ so that $d_\cD(\phi_{(t,u]},\id)<\ve/2$ for all $u\in(t,t+\d]$.

1406: For such $u$ and for $x$ a point of continuity of $\phi_{(s,t]}$, we have

1407: $$

1408: \phi_{(s,t]}^+(x)-\ve=\phi_{(s,t]}^-(x)-\ve\leq\phi_{(t,u]}^-\circ\phi_{(s,t]}^-(x)\leq 

1409: \phi_{(s,u]}^-(x)\leq 

1410: \phi_{(s,u]}^+(x) \leq \phi_{(t,u]}^+ \circ \phi_{(s,t]}^+(x) \leq \phi_{(s,t]}^+(x) +  \ve,

1411: $$

1412: so $|\phi_{(s,u]}^+(x)-\phi_{(s,t]}^+(x)|\leq\ve$. The final estimate extends to all $x$ by right-continuity.

1413: Hence the map is right continuous. A similar argument shows that, for $u\in(s,t)$, we have

1414: $|\phi_{(s,u]}^+(x)-\phi_{(s,t)}^+(x)|\to0$ as $u\to t$, so that the map has a left limit at $t$

1415: given by $\phi_{(s,t)}^+(x)$. Finally, if $\phi\in C^\circ(\R,\cD)$, then

1416: $\phi_{(s,t)}=\phi_{(s,t]}$, so the map is continuous.

1417: \end{proof}

1418: 

1419: \begin{proposition}\label{PHICONT}

1420: For all $\phi\in C^\circ(\R,\cD)$, the map $(s,t)\mapsto\phi_{ts}:\{(s,t):s\le t\}\to\cD$ is continuous.

1421: Moreover, for all $\phi\in D^\circ(\R,\cD)$ and for any sequence of finite intervals $I_n\to I$, we have $\phi_{I_n}\to\phi_I$.

1422: \end{proposition}

1423: \begin{proof}

1424: The first assertion follows from the second: given $\phi\in C^\circ(\R,\cD)$ and sequences $s_n\to s$ and $t_n\to t$, then,

1425: passing to a subsequence if necessary, we can assume that $(s_n,t_n]\to I$ for some interval $I$ with $\inf I=s$ and $\sup I=t$.

1426: Then, by the second assertion, we have $\phi_{t_ns_n}\to\phi_I=\phi_{ts}$, as required.

1427: 

1428: So, let us fix $\phi\in D^\circ(\R,\cD)$ and a sequence of finite intervals $I_n\to I$.

1429: By combining the cadlag and weak flow properties, we can show the following variant of the cadlag property: for all $t\in\R$, we have 

1430: \begin{equation}\label{WFP2}

1431: \phi_{[s,t)}\to\id\q\text{as $s\uparrow t$},\q

1432: \phi_{(t,u]}\to\id\q\text{as $u\downarrow t$}.

1433: \end{equation}

1434: For each $n$, there exist two disjoint intervals $J_n$ and $J_n'$, possibly empty,

1435: such that $I\triangle I_n=J_n\cup J_n'$. For any such $J_n$ and $J_n'$, using the weak flow property, we obtain

1436: $$

1437: d_\cD(\phi_I,\phi_{I_n})\le\|\phi_{J_n}-\id\|+\|\phi_{J_n'}-\id\|.

1438: $$

1439: Set $s=\inf I$, $s_n=\inf I_n$, $t=\sup I$ and $t_n=\sup I_n$.

1440: Then $s_n\to s$, $t_n\to t$, and

1441: $$

1442: \text{ if $s\in I$ then $s\in I_n$ eventually},\q

1443: \text{ if $s\not\in I$ then $s\not\in I_n$ eventually},

1444: $$$$

1445: \text{ if $t\in I$ then $t\in I_n$ eventually},\q

1446: \text{ if $t\not\in I$ then $t\not\in I_n$ eventually}.

1447: $$

1448: Hence, using the cadlag property or (\ref{WFP2}), or both, we find that

1449: $\phi_{J_n}\to\id$ and $\phi_{J_n'}\to\id$, which proves the proposition.

1450: \end{proof}

1451: 

1452: 

1453: \begin{proposition}

1454: The metrics $d_C$ and $d_D$ generate the same topology on $C^\circ(\R,\cD)$.

1455: \end{proposition}

1456: \begin{proof}

1457: On comparing the definitions of $d^C_n$ and $d^D_n$ for each $n\in\N$, and considering the choice $\l=\id$,

1458: we see that $d_D\le d_C$. Hence, it will suffice to show, given $\phi\in C^\circ(\R,\cD)$, $n\in\N$ and $\ve>0$,

1459: that there exists $\ve'>0$ such that, for all $\psi\in C^\circ(\R,\cD)$, we have $d_C^{(n)}(\phi,\psi)<\ve$

1460: whenever $d_{n+1}^D(\phi,\psi)<\ve'$. By the preceding proposition, there exists a $\d\in(0,1]$ such that

1461: $d_\cD(\phi_{ts},\phi_{t's'})<\ve/2$ whenever $|s-s'|,|t-t'|\le\d$ and $s,t\in(-n,n)$. Set $\ve'=\d\wedge(\ve/2)$

1462: and suppose that $d_{n+1}^D(\phi,\psi)<\ve'$. Then there exists an increasing homeomorphism $\l$ of $\R$, with

1463: $|\l(t)-t|\le\d$ for all $t$, such that, for all intervals $I$, we have 

1464: $\|\chi_{n+1}(I)\psi_I^\times-\chi_{n+1}(\l(I))\phi_{\l(I)}^\times\|<\ve/2$.

1465: Given $s,t\in(-n,n)$ with $s<t$, take $I=(s,t]$.

1466: Then $\chi_{n+1}(I)=\chi_{n+1}(\l(I))=1$, so $d_\cD(\phi_{\l(t)\l(s)},\psi_{ts})=\|\psi_I^\times-\phi_{\l(I)}^\times\|<\ve/2$.

1467: But then, for all such $s,t$, we have 

1468: $$

1469: d_\cD(\phi_{ts},\psi_{ts})\le d_\cD(\phi_{ts},\phi_{\l(t)\l(s)})+d_\cD(\phi_{\l(t)\l(s)},\psi_{ts})<\ve,

1470: $$

1471: so $d_C^{(n)}(\phi,\psi)<\ve$, as required.

1472: \end{proof}

1473: 

1474: 

1475: \begin{proposition}

1476: The metric spaces

1477: $(C^\circ(\R,\cD),d_C)$ and $(D^\circ(\R,\cD),d_D)$ are complete and separable.

1478: \end{proposition}

1479: \begin{proof}

1480: The argument for completeness is a variant of the corresponding

1481: argument for the usual Skorokhod space $D(\R, S)$ of cadlag paths in complete separable metric

1482: space $S$, as found for example in \cite{B}.

1483: Suppose then that $(\psi^n)_{n \geq 1}$ is a Cauchy sequence in $D^\circ(\R,\cD)$.

1484: There exists a subsequence $\phi^k=\psi^{n_k}$ such that $d_D^{(n)}(\phi^n, \phi^{n+1}) < 2^{-n}$ for all $n\ge1$.

1485: It will suffice to find a limit in $D^\circ(\R,\cD)$ for $(\phi^n)_{n \geq 1}$.

1486: There exist increasing homeomorphisms $\mu_n$ of $\R$

1487: for which $\gamma(\mu_n) < 2^{-n}$ and

1488: $$

1489: d_\cD(\phi^n_I, \phi^{n+1}_{\mu_n(I)})<2^{-n},\q I\cup\mu_n(I)\sse(-n,n).

1490: $$

1491: For each $n\ge1$, the sequence $(\mu_{n+m} \circ \cdots \circ \mu_{n})_{m\ge1}$ converges

1492: uniformly on $\R$ to an increasing homeomorphism, $\lambda_n$ say,

1493: with $\gamma(\lambda_n) < 2^{-n+1}$. Then $\mu_n\circ\l_n^{-1}=\l_{n+1}^{-1}$, so

1494: $$

1495: d_\cD(\phi^n_{\lambda_n^{-1}(I)}, \phi^{n+1}_{\lambda_{n+1}^{-1}(I)}) < 2^{-n},\q

1496: I\sse(-n+1,n-1).

1497: $$

1498: So, for all $m\ge n$,

1499: \begin{equation}\label{PHMN}

1500: d_\cD(\phi^n_{\lambda_n^{-1}(I)}, \phi^{n+m}_{\lambda_{n+m}^{-1}(I)}) < 2^{-n+1},\q I\sse(-n+1,n-1).

1501: \end{equation}

1502: Hence, for all finite intervals $I\sse\R$,

1503: $(\phi^n_{\lambda_n^{-1}(I)})_{n \geq 1}$ is a

1504: Cauchy sequence in $\cD$, which, since $\cD$ is complete, has a limit $\phi_I\in\cD$.

1505: On letting $m\to\infty$ in (\ref{PHMN}), we obtain

1506: $$

1507: d_\cD(\phi^n_{\lambda_n^{-1}(I)},\phi_I) < 2^{-n+1},\q I\sse(-n+1,n-1).

1508: $$

1509: By Proposition \ref{WFL}, $\phi=(\phi_I:I\sse\R)$ has the weak flow property.

1510: To see that $\phi$ is cadlag, suppose given $\ve>0$ and $t\in\R$. Choose $n$ such that

1511: $2^{-n+1}\le\ve/3$ and $|t|\le n-2$. Then choose $\d\in(0,1]$ such that

1512: $$

1513: d_\cD(\phi^n_{\l_n^{-1}(s,t)},\id)<\ve/3,\q d_\cD(\phi^n_{\l_n^{-1}(t,u)},\id)<\ve/3

1514: $$

1515: whenever $s\in(t-\d,t)$ and $u\in(t,t+\d)$. For such $s$ and $u$, we then have

1516: $$

1517: d_\cD(\phi_{(s,t)},\id)<\ve,\q d_\cD(\phi_{(t,u)},\id)<\ve.

1518: $$

1519: Hence $\phi\in D^\circ(\R,\cD)$. For $m\le n-3$, we have

1520: \begin{align*}

1521: d_D^{(m)}(\phi^n,\phi)

1522: &\le\g(\l_n)\vee\sup_{I\sse(-m-2,m+2)}\|\chi_m(\l_n^{-1}(I))\phi_{\l_n^{-1}(I)}^{n\times}-\chi_m(I)\phi_I^\times\|\\

1523: &\le\g(\l_n)\vee\sup_{I\sse(-m-2,m+2)}\left\{d_\cD(\phi^n_{\lambda_n^{-1}(I)},\phi_I)+\g(\l_n)\|\phi_I^\times\|\right\}\\

1524: &\le2^{-n+1}(1+\sup_{I\sse(-m-2,m+2)}\|\phi_I^\times\|).

1525: \end{align*} 

1526: Hence $d_D(\phi^n,\phi)\to0$ as $n\to\infty$.

1527: We have shown that $D^\circ(\R,\cD)$ is complete.

1528: If the sequence $(\phi^n)_{n \geq 1}$ in fact lies in $C^\circ(\R,\cD)$, then,

1529: by an obvious variation of the argument for the cadlag property, the limit

1530: $\phi$ also lies in $C^\circ(\R,\cD)$. Hence $C^\circ(\R,\cD)$ is also complete.

1531: In particular, $C^\circ(\R,\cD)$ is a closed subspace in $D^\circ(\R,\cD)$.

1532: 

1533: We turn to the question of separability.

1534: Let us write $D_N$ for the set of those $\phi\in D^\circ(\R,\cD)$ such that:

1535: \begin{itemize}

1536: \item[(i)] for some $n\in\N$ and some rationals $t_1<\dots<t_n$, we have

1537: $\phi_J=\id$ for all time intervals $J$ which do not intersect the set $\{t_1,\dots,t_n\}$;

1538: \item[(ii)] for all other time intervals $I$, the maps $\phi_I$ and $\phi_I^{-1}$ on $\R$

1539: are constant on all space intervals which do not intersect $2^{-N}\Z$.

1540: \end{itemize}

1541: Note that each $\phi\in D_N$ is determined by the maps $\phi_{(t_k,t_m]}$,

1542: for integers $0\le k<m\le n$, where $t_0<t_1$,

1543: and for each of these maps there are only countably many possibilities (finitely many

1544: if we insist that $\phi(0)\in[0,1)$).

1545: Hence $D_N$ is countable and so is $D_*=\bigcup_{N\ge1}D_N$.

1546: We shall show that $D_*$ is also dense in $D^\circ(\R,\cD)$.

1547: 

1548: Fix $\phi\in D^\circ(\R,\cD)$ and $n_0\ge1$. It will suffice to find, for a given $\ve>0$,

1549: a $\psi\in D_*$ with $d^{(n_0)}_D(\phi,\psi)<\ve$. By the cadlag property and compactness,

1550: there exist $n\in\N$ and reals $s_1<\dots<s_n$ in $I_0=(-n_0-1,n_0+1)$ such that

1551: $d_\cD(\phi_I,\id)<\ve/4$ for every subinterval $I$ of $I_0$ which does not intersect $\{s_1,\dots,s_n\}$.

1552: Then we can find rationals $t_1<\dots<t_n$ in $I_0$ and

1553: an increasing homeomorphism $\l$ of $\R$, with $\l(t)=t$ for $t\not\in I_0$, with $\g(\l)\sup_{I\sse I_0}\|\phi^\times_I\|<\ve/4$,

1554: and such that $\l(t_m)=s_m$ for all $m$.

1555: Set $s_0=t_0=-n_0-1$.

1556: 

1557: For $f\in\cD$, write $\Delta(f)$ for the set of points where $f$ is not continuous.

1558: Define, for $m=0,1,\dots,n$,

1559: $$

1560: \Delta_m=\bigcup_{k=0}^{m-1}\Delta(\phi_{(s_k,s_m]}^{-1})\cup\bigcup_{k=m+1}^n\Delta(\phi_{(s_m,s_k]}).

1561: $$

1562: Then $\Delta_m$ is countable, so we can choose $N\ge1$ with $16.2^{-N}\le\ve$ and choose

1563: $\ve_m\in\R$ with $|\ve_m|\le2^{-N}$ such that

1564: $$

1565: \t_m(\Delta_m)\cap2^{-N}\Z=\es,\q m=0,1,\dots,n,

1566: $$

1567: where $\t_m(x)=x+\ve_m$. Set

1568: $$

1569: \d^-(x)=2^N\lceil2^{-N}x\rceil,\q \d^+(x)=2^N\lfloor2^{-N}x\rfloor+1.

1570: $$

1571: Note that $\d=\{\d^-,\d^+\}\in\cD$. Define for $0\le k<m\le n$

1572: $$

1573: \psi_{(t_k,t_m]}^-=(\d^{-1})^-\circ(\t_m)^{-1}\circ\phi_{(s_k,s_m]}^-\circ\t_k\circ\d^-,\q

1574: \psi_{(t_k,t_m]}^+=(\d^{-1})^+\circ(\t_m)^{-1}\circ\phi_{(s_k,s_m]}^+\circ\t_k\circ\d^+.

1575: $$

1576: Then $\psi_{(t_k,t_m]}=\{\psi_{(t_k,t_m]}^-,\psi_{(t_k,t_m]}^+\}\in\cD$

1577: by our choice of $\ve_k$ and $\ve_m$.

1578: Moreover $\d^+\circ(\d^{-1})^+\ge\id$ and $\d^-\circ(\d^{-1})^-\le\id$ so,

1579: for $0\le m<m'<m''\le n$, we obtain the inequalities

1580: $$

1581: \psi_{(t_{m'},t_{m''}]}^-\circ\psi_{(t_m,t_{m'}]}^-\le\psi_{(t_m,t_{m''}]}^-

1582: \le\psi_{(t_m,t_{m''}]}^+\le\psi_{(t_{m'},t_{m''}]}^+\circ\psi_{(t_m,t_{m'}]}^+

1583: $$

1584: from the corresponding inequalities for $\phi$.

1585: We use the equations $\|\d-\id\|=2^{-N}$ and $\|\t_m-\id\|=|\ve_m|$

1586: to see that

1587: $$

1588: d_\cD(\phi_{(s_k,s_m]},\psi_{(t_k,t_m]})\le4.2^{-N},\q 0\le k<m\le n.

1589: $$

1590: Define $\psi_J=\psi_{(t_k,t_m]}$ for all intervals $J$ such that

1591: $J\cap\{t_1,\dots,t_n\}=\{t_{k+1},\dots,t_m\}$.

1592: For such intervals $J$, with $J\sse I_0$, we have

1593: $d_\cD(\phi_{(s_k,s_m]\sm\l(J)},\id)<\ve/4$ and $d_\cD(\phi_{\l(J)\sm(s_k,s_m]},\id)<\ve/4$ so,

1594: using the weak flow property for $\phi$,

1595: $$

1596: d_\cD(\psi_J, \phi_{\l(J)})\le d_\cD(\psi_{(t_k,t_m]},\phi_{(s_k,s_m]})

1597: +d_\cD(\phi_{(s_k,s_m]},\phi_{\l(J)})\le 4.2^{-N}+2\ve/4<3\ve/4.

1598: $$

1599: Define $\psi_J=\id$ for all intervals $J$ which do not intersect $\{t_1,\dots,t_n\}$.

1600: For such intervals $J$ with $J\sse I_0$, we have $d_\cD(\psi_J, \phi_{\l(J)})\le d_\cD(\id,\phi_{\l(J)})\le\ve/4$.

1601: Now $\psi\in D_N$ and 

1602: $$

1603: d^{(n_0)}_D(\phi,\psi)\le\g(\l)\vee\sup_{J\sse I_0}\{d_\cD(\psi_J,\phi_{\l(J)})+\g(\l)\|\phi_J^\times\|\}<\ve, 

1604: $$

1605: as required.

1606: This proves that $D^\circ(\R,\cD)$ is separable and, since $C^\circ(\R,\cD)$

1607: is a closed subspace of $D^\circ(\R,\cD)$, it follows that $C^\circ(\R,\cD)$

1608: is also separable.

1609: \end{proof}

1610: 

1611: \begin{proposition}

1612: For all $s,t\in\R$ with $s<t$, and all $x\in\R$, the map

1613: $\phi\mapsto\phi_{ts}^+(x)$ on $C^\circ(\R,\cD)$ is continuous.

1614: Moreover the Borel $\s$-algebra on $C^\circ(\R,\cD)$ is generated by

1615: the set of all such maps with $s,t$ and $x$ rational.

1616: 

1617: For all finite intervals $I\sse\R$ and all $x\in\R$, the 

1618: map $\phi\mapsto\phi_I^+(x)$ on $D^\circ(\R,\cD)$ is Borel 

1619: measurable, and the Borel $\s$-algebra on $D^\circ(\R,\cD)$ is generated by

1620: the set of all such maps where $I=(s,t]$ with $s,t$ and $x$ rational.

1621: \end{proposition}

1622: \begin{proof}

1623: The results for $C^\circ(\R,\cD)$ can be proved more simply than those for $D^\circ(\R,\cD)$.

1624: We omit details of the former, but note that these follow also from the latter, by general 

1625: measure theoretic arguments, given what we already know about the two spaces. 

1626: 

1627: The proof for $D^\circ(\R,\cD)$ is an adaptation of the analogous result for the classical Skorokhod 

1628: space, see for example \cite[page 335]{MR1943877}.

1629: We prove first the Borel measurability of the evaluation maps.

1630: Given a finite interval $I$ and $x\in\R$, we can find $s_n,t_n\in\R$ such that $(s_n,t_n]\to I$ as $n\to\infty$.

1631: Then $\phi^+_I(x)=\lim_{m\to\infty}\lim_{n\to\infty}\phi^+_{(s_n,t_n]}(x+1/m)$, by Proposition \ref{PHICONT}.

1632: Hence, it will suffice to consider intervals $I$ of the form $(s,t]$. Fix $s,t$ and $x$ and define for each $m,n\in\N$

1633: a function $F_{m,n}$ on $D^\circ(\R,\cD)$ by 

1634: $$

1635: F_{m,n}(\phi)=\int_s^{s+1/n}\int_t^{t+1/n}\int_x^{x+1/m}\phi^+_{(s',t']}(x')dx'dt'ds'.

1636: $$

1637: Suppose $\phi^k\to\phi$ in $D^\circ(\R,\cD)$. We can choose increasing homeomorphisms $\l_k$ of $\R$ such that,

1638: $\g(\l_k)\to0$ and, uniformly in $r\in[s-1,s+1]$ and $u\in[t-1,t+1]$, we have

1639: $$

1640: d_\cD(\phi^k_{\l_k(r,u]},\phi_{(r,u]})\to0.

1641: $$

1642: Define

1643: $$

1644: f(r,u)=\int_x^{x+1/m}\phi_{\l(r,u]}(x')dx',\q

1645: f_k(r,u)=\int_x^{x+1/m}\phi^k_{\l_k(r,u]}(x')dx'.

1646: $$

1647: Then $f_k(r,u)\to f(r,u)$, uniformly in $r\in[s-1,s+1]$ and $u\in[t-1,t+1]$.

1648: Set $\mu_k=\l_k^{-1}$. Then

1649: \begin{align*}

1650: F_{m,n}(\phi^k)&=\int_{\mu_k(s)}^{\mu_k(s+1/n)}\int_{\mu_k(t)}^{\mu_k(t+1/n)}f_k(r,u)d\l_k(u)d\l_k(r)\\

1651: &\to\int_s^{s+1/n}\int_t^{t+1/n}f(r,u)dudr= F_{m,n}(\phi),

1652: \end{align*}

1653: so $F_{m,n}$ is continuous on $D^\circ(\R,\cD)$. By Proposition \ref{PHICONT}, we have 

1654: $$

1655: \phi_{(s,t]}^+(x)=\lim_{m\to\infty}\lim_{n\to\infty}\frac1{mn^2}F_{m,n}(\phi).

1656: $$

1657: Hence $\phi\mapsto\phi_{(s,t]}^+(x)$ is Borel measurable, as required.

1658: 

1659: Write now $\cE$ for the $\s$-algebra on $D^\circ(\R,\cD)$ generated by all maps of this form

1660: with $s,t$ and $x$ rational. 

1661: It remains to show that $\cE$ contains the Borel $\s$-algebra of $D^\circ(\R,\cD)$.

1662: Write $\{(I_k,z_k):k\in\N\}$ for an enumeration of the set $\{(s,t]:s,t\in\Q,s<t\}\times\Q$.

1663: It is straightforward to show that, for all $k$, the map $\phi\mapsto\phi_{I_k}^\times(z_k)$ is $\cE$-measurable.

1664: Fix $n\in\N$, $\phi^0\in D^\circ(\R,\cD)$, $r\in(0,\infty)$ and $k\in\N$, and consider the set

1665: $$

1666: A(k,r)=\{\phi\in D^\circ(\R,\cD):(\chi_n(I_1)\phi_{I_1}^\times(z_1),\dots,\chi_n(I_k)\phi_{I_k}^\times(z_k))\in B(k,r)\},

1667: $$

1668: where

1669: $$

1670: B(k,r)=\bigcup_\l\{(y_1,\dots,y_k)\in\R^k:\max_{j\le k}|y_j-\chi_n(\l(I_j))\phi_{\l(I_j)}^{0\times}(z_j)|<r\},

1671: $$

1672: where the union is taken over all increasing homeomorphisms $\l$ of $\R$ with $\g(\l)<r$.

1673: Note that $B(k,r)$ is an open set in $\R^k$, so $A(k,r)\in\cE$, so $A=\bigcup_{m\in\N}\bigcap_{k\in\N}A(k,r-1/m)\in\cE$.

1674: 

1675: Consider the set 

1676: $$

1677: C=\{\phi\in D^\circ(\R,\cD):d_D^{(n)}(\phi,\phi^0)<r\}.

1678: $$

1679: It is straightforward to check from the definition of $d_D^{(n)}$, that $C\sse A$. Suppose that $\phi\in A$. We shall

1680: show that $\phi\in C$. Then $C=A$, so $C\in\cE$, and since sets of this form generate the Borel $\s$-algebra, we are done.

1681: 

1682: We can find an $m\in\N$ and, for each $k\in\N$, a $\l_k$ with $\g(\l_k)<r-1/m$ such that 

1683: $$

1684: |\chi_n(I_j)\phi_{I_j}^\times(z_j)-\chi_n(\l_k(I_j))\phi_{\l_k(I_j)}^{0\times}(z_j)|<r-1/m,\q j=1,\dots,k.

1685: $$

1686: Without loss of generality, we may assume that the sequence $(\l_k:k\in\N)$ converges uniformly on compacts, and 

1687: that its limit, $\l$ say, satisfies $\g(\l)\le r-1/m$.

1688: By Proposition \ref{PHICONT}, for each $j$, there is an interval $\hat I_j$, having the same endpoints as $I_j$

1689: such that $\phi_{\l(\hat I_j)}$ is a limit point in $\cD$ of the sequence $(\phi_{\l_k(I_j)}:k\in\N)$,

1690: so $\phi_{\l(\hat I_j)}^\times$ is a limit point in $\cS$ of the sequence $(\phi_{\l_k(I_j)}^\times:k\in\N)$. Then

1691: $$

1692: |\chi_n(I_j)\phi_{I_j}^\times(z_j)-\chi_n(\l(\hat I_j))\phi_{\l(\hat I_j)}^{0\times}(z_j)|\le r-1/m,

1693: $$

1694: for all $j$. For all finite intervals $I$ and all $z\in\R$, we can find a sequence $(j_p:p\in\N)$ such that

1695: $I_{j_p}\to I$, $\hat I_{j_p}\to I$ and $z_{j_p}\to z$. So we obtain

1696: $$

1697: |\chi_n(I)\phi_{I}^\times(z)-\chi_n(\l(I))\phi_{\l(I)}^{0\times}(z)|\le r-1/m.

1698: $$

1699: Hence $d_D^{(n)}(\phi,\phi^0)\le r-1/m$ and $\phi\in C$, as we claimed.

1700: \end{proof}

1701: 

1702: Recall that we define, for $e=(s,x)\in\R^2$ and $\phi\in D^\circ(\R,\cD)$,

1703: $$

1704: Z^{e,+}(\phi)=(\phi^+_{(s,t]}(x):t\ge s)

1705: $$

1706: and for $E\sse\R^2$, we define $Z^{E,+}:D^\circ(\R,\cD)\to D_E$ by $Z^{E,+}(\phi)^e=Z^{e,+}(\phi)$.

1707: Recall also that we set

1708: $$

1709: C^{\circ,+}_E=\{Z^{E,+}(\phi):\phi\in C^\circ(\R,\cD)\},\q

1710: D^{\circ,+}_E=\{Z^{E,+}(\phi):\phi\in D^\circ(\R,\cD)\},

1711: $$

1712: and that we define analogously $Z^{E,-}$, $C^{\circ,-}_E$ and $D^{\circ,-}_E$, 

1713: and we set $C_E^\circ=C^{\circ,+}_E\cap C^{\circ,-}_E$ and $D_E^\circ=D^{\circ,+}_E\cap D^{\circ,-}_E$.

1714: 

1715: \begin{proposition}

1716: Let $E$ be a countable subset of $\R^2$ containing $\Q^2$.

1717: Then $Z^{E,+}:C^\circ(\R,\cD)\to C^{\circ,+}_E$ is a bijection,

1718: $C^{\circ,+}_E$ is a measurable subset of $C_E$, and

1719: the inverse bijection $\Phi^{E,+}:C^{\circ,+}_E\to C^\circ(\R,\cD)$ is a measurable map. 

1720: 

1721: Moreover $Z^{E,+}:D^\circ(\R,\cD)\to D^{\circ,+}_E$ is also a bijection,

1722: $D^{\circ,+}_E$ is a measurable subset of $D_E$

1723: and the inverse bijection $\Phi^{E,+}:D^{\circ,+}_E\to D^\circ(\R,\cD)$ is also a measurable map. 

1724: 

1725: Moreover the same statements hold with $+$ replaced by $-$, and we have $\Phi^{E,+}=\Phi^{E,-}$ on $D_E^\circ$.

1726: \end{proposition}

1727: \begin{proof}

1728: We discuss only the cadlag case. The usual comments apply concerning the relationship of this

1729: case with the continuous case. The simple argument that $Z^{E,+}$ is injective on $D^{\circ,+}_E$ 

1730: was given in the proof of Theorem \ref{MAIN}.

1731: Consider for $z\in D_E$ the conditions

1732: \begin{equation}\label{ZDEG1}

1733: z_t^{(s,x+n)}=z_t^{(s,x)}+n,\q s,t,x\in\Q,\q s<t,\q n\in\Z

1734: \end{equation}

1735: and

1736: \begin{equation}\label{ZRIGHT}

1737: z_t^{(s,x)}=\inf_{y\in\Q,y>x}z_t^{(s,y)},\q (s,x)\in E,\q t\in\Q,t>s.

1738: \end{equation}

1739: Under these conditions, define for $s,t\in\Q$ with $s<t$ and for $x\in\R$,

1740: $$

1741: \Phi_{(s,t]}^-(x)=\sup_{y\in\Q,y<x}z_t^{(s,y)},\q

1742: \Phi_{(s,t]}^+(x)=\inf_{y\in\Q,y>x}z_t^{(s,y)}.

1743: $$

1744: Then $\Phi_{(s,t]}=\{\Phi_{(s,t]}^-,\Phi_{(s,t]}^+\}\in\cD$ and

1745: $$

1746: \Phi_{(s,t]}^+(x)=z_t^{(s,x)},\q s,t,x\in\Q,\q s<t.

1747: $$

1748: Now consider the following additional conditions on $z$: 

1749: \begin{equation}\label{ZFLOW}

1750: \Phi_{(t,u]}^-\circ\Phi_{(s,t]}^-\le\Phi_{(s,u]}^-\le\Phi_{(s,u]}^+\le\Phi_{(t,u]}^+\circ\Phi_{(s,t]}^+,\q s,t,u\in\Q,\q s<t<u;

1751: \end{equation}

1752: and

1753: \begin{itemize}

1754: \item[]for all $\ve>0$ and all $n\in\N$, there exist $\d>0$, $m\in\Z^+$ and $u_1,\dots,u_m\in(-n,n)$ such that

1755: \end{itemize}

1756: \begin{equation}\label{ZCADLAG}

1757: \|\Phi_{(s,t]}-\id\|<\ve 

1758: \end{equation}

1759: \begin{itemize}

1760: \item[]whenever $s,t\in\Q\cap(-n,n)$ with $0<t-s<\d$ and $(s,t]\cap\{u_1,\dots,u_m\}=\es$.

1761: \end{itemize}

1762: Note that the inequalities between functions required in (\ref{ZFLOW}) hold whenever the same inequalities

1763: hold between their restrictions to $\Q$, by left and right continuity.

1764: Note also that condition (\ref{ZCADLAG}) is equivalent to the following condition involving quantifiers only over countable sets:

1765: \begin{itemize}

1766: \item[]for all rationals $\ve>0$ and all $n\in\N$, there exist a rational $\d>0$ and an $m\in\Z^+$ such that, for all rationals $\eta>0$,

1767: there exist rationals $s_1,t_1,\dots,s_m,t_m\in(-n,n)$, with $s_i<t_i$ for all $i$ and with

1768: $\sum_{i=1}^m(t_i-s_i)<\eta$, such that

1769: \end{itemize}

1770: $$

1771: \|\Phi_{(s,t]}-\id\|<\ve

1772: $$

1773: \begin{itemize}

1774: \item[]whenever $s,t\in\Q\cap(-n,n)$ with $0<t-s<\d$ and $(s,t]\cap((s_1,t_1]\cup\dots\cup(s_m,t_m])=\es$.

1775: \end{itemize}

1776: Denote by $D_E^{*,+}$ the set of those $z\in D_E$ where conditions (\ref{ZDEG1}), (\ref{ZRIGHT}), (\ref{ZFLOW}) and (\ref{ZCADLAG})

1777: all hold. Then $D^{*,+}_E$ is a measurable subset of $D_E$. Fix $z\in D^{*,+}_E$. Given finite interval $I$, we can find

1778: sequences of rationals $s_n$ and $t_n$ such that $(s_n,t_n]\to I$ as $n\to\infty$. Then, by conditions (\ref{ZFLOW}) and (\ref{ZCADLAG}),

1779: $$

1780: d_\cD(\Phi_{(s_n,t_n]},\Phi_{(s_m,t_m]})\le\|\Phi_{(s_n,s_m]}-\id\|+\|\Phi_{(t_n,t_m]}-\id\|\to0

1781: $$

1782: as $n,m\to\infty$. So the sequence $\Phi_{(s_n,t_n]}$ converges in $\cD$, with limit $\Phi_I$, say, and $\Phi_I$ does not

1783: depend on the approximating sequences of rationals.

1784: In the case where $I=I_1\oplus I_2$, there exists another sequence of rationals $u_n$ such that $(s_n,u_n]\to I_1$

1785: and $(u_n,t_n]\to I_2$ as $n\to\infty$. Hence $\Phi=(\Phi_I:I\sse\R)$ has the weak flow property, by Proposition

1786: \ref{WFL}. It is straightforward to deduce from (\ref{ZCADLAG}) that $\Phi$ is moreover cadlag, so $\Phi=\Phi(z)\in D^\circ(\R,\cD)$.

1787: It follows from its construction, and the preceding proposition, that the map $z\mapsto\Phi(z):D^{*,+}_E\to D^\circ(\R,\cD)$ is measurable.

1788: 

1789: Now, for all $z\in D^{*,+}_E$, we have $Z^{E,+}(\Phi(z))=z$ and for all $\phi\in D^\circ(\R,\cD)$, we have $Z^{E,+}(\phi)\in D^*_E$ and

1790: $\Phi(Z^{E,+}(\phi))=\phi$. Hence $D_E^{\circ,+}=D^{*,+}_E$ and $\Phi^{E,+}=\Phi$, showing that these are measurable, as claimed.

1791: 

1792: Consider now for $z\in D_E$ the condition

1793: \begin{equation}\label{ZLEFT}

1794: z_t^{(s,x)}=\sup_{y\in\Q,y<x}z_t^{(s,y)},\q (s,x)\in E,\q t\in\Q, t>s.

1795: \end{equation}

1796: Denote by $D_E^{*,-}$ the set of those $z\in D_E$ where conditions (\ref{ZDEG1}), (\ref{ZFLOW}), (\ref{ZCADLAG}) and 

1797: (\ref{ZLEFT}) all hold, and define $\Phi$ on $D^{*,-}_E$ exactly as on $D^{*,+}_E$. 

1798: Then, by a similar argument, $D_E^{\circ,-}=D^{*,-}_E$ and $\Phi=\Phi^{E,-}$ on $D_E^{\circ,-}$.

1799: In particular $\Phi^{E,+}=\Phi^{E,-}$ on $D_E^\circ$, as claimed.

1800: \end{proof}

1801: 

1802: \begin{proposition}\label{MPI}

1803: Let $E$ be a countable subset of $\R^2$ containing $\Q^2$.

1804: Then $\mu_E(C^\circ_E)=1$.

1805: \end{proposition}

1806: \begin{proof}

1807: We use an identification of $C^\circ_E$ analogous to that implied for $D^\circ_E$ by the preceding proof.

1808: The same five conditions (\ref{ZDEG1}), (\ref{ZRIGHT}), (\ref{ZFLOW}), (\ref{ZCADLAG}) and (\ref{ZLEFT}) characterize

1809: $C^\circ_E$ inside $C_E$, except that, in (\ref{ZCADLAG}), only the case $m=0$ is allowed.

1810: Recall that, under $\mu_E$, for time-space starting points $e=(s,x)$ and $e'=(s',x')$, 

1811: the coordinate processes $Z^e$ and $Z^{e'}$ behave as independent Brownian motions up to 

1812: $$

1813: T^{ee'}=\inf\{t\ge s\vee s':Z_t^e-Z_t^{e'}\in\Z\},

1814: $$

1815: after which they continue to move as Brownian motions, but now with a constant separation.

1816: In particular, if $s=s'$ and $x'=x+n$ for some $n\in\Z$, then $T^{ee'}=0$, so $Z^{e'}_t=Z^e_t+n$

1817: for all $t\ge s$, so (\ref{ZDEG1}) holds almost surely.

1818: 

1819: Let $(s,x)\in E$ and $t,u\in\Q$, with $s\le t<u$.

1820: Consider the event

1821: $$

1822: A=\left\{\sup_{y\in\Q,y<Z_t^{(s,x)}}Z_u^{(t,y)}=Z_u^{(s,x)}=\inf_{y'\in\Q,y'>Z_t^{(s,x)}}Z_u^{(t,y')}\right\}.

1823: $$

1824: Fix $n\in\N$ and set $Y=n^{-1}\lfloor nZ_t^{(s,x)}\rfloor$ and $Y'=Y+1/n$.

1825: Then $Y$ and $Y'$ are $\cF_t$-measurable, $\Q$-valued random variables.

1826: Now $\PP(Y<Z_t^{(s,x)}<Y')=1$ and 

1827: $$

1828: \{Y<Z_t^{(s,x)}<Y'\}\cap\{T^{(t,Y)(t,Y')}\le u\}\sse A.

1829: $$

1830: By the Markov property of Brownian motion, almost surely,

1831: $$

1832: \PP(T^{(t,Y)(t,Y')}\le u|\cF_t)=2\Phi\left(\frac1{n\sqrt{2(u-t)}}\right),

1833: $$

1834: and the right-hand side tends to $1$ as $n\to\infty$.

1835: So, by bounded convergence, we obtain $\PP(A)=1$. 

1836: On taking a countable intersection of such sets $A$ over the possible values

1837: of $s,x,t$ and $u$, we deduce that

1838: conditions (\ref{ZRIGHT}), (\ref{ZFLOW}) and (\ref{ZLEFT}) hold almost surely.

1839: 

1840: It remains to establish the continuity condition (\ref{ZCADLAG}). 

1841: For a standard Brownian motion $B$ starting from $0$, we have, for $n\ge4$,

1842: $$

1843: \PP\left(\sup_{t\le1}|B_t|>n\right)\le e^{-n^2/2}.

1844: $$

1845: Define, for $\d>0$ and $e=(s,x)\in E$, 

1846: $$

1847: V^e(\d)=\sup_{s\le t\le s+\d^2}|Z^e_t-x|.

1848: $$

1849: Then, by scaling, 

1850: $$

1851: \PP(V^e(\d)>n\d)\le e^{-n^2/2}.

1852: $$

1853: Consider, for each $n\in\N$ the set 

1854: $$

1855: E_n=\{(j2^{-2n},k2^{-n}):j\in\tfrac12\Z\cap[-2^{2n},2^{2n}),k=0,1,\dots,2^n-1\}

1856: $$

1857: and the event

1858: $$

1859: A_n=\bigcup_{e\in E_n}\{V^e(2^{-n})>n2^{-n}\}.

1860: $$

1861: Then $\PP(A_n)\le|E_n|e^{-n^2/2}$, so $\sum_n\PP(A_n)<\infty$, so by Borel--Cantelli, almost surely, 

1862: there is a random $N<\infty$ such that $V^e(2^{-n})\le n2^{-n}$ for all $e\in E_n$, for all $n\ge N$.

1863: 

1864: Given $\ve>0$, choose $n\ge N$ such that $(4n+2)2^{-n}\le\ve$ and set $\d=2^{-2n-1}$. Then, for all

1865: rationals $s,t\in(-n,n)$ with $0<t-s<\d$ and all rationals $x\in[0,1]$, there exist $e^\pm=(r,y^\pm)\in E_n$ such that 

1866: $$

1867: r\le s<t\le r+2^{-2n},\q  x+n2^{-n}<y^+\le x+(n+1)2^{-n},\q x-(n+1)2^{-n}\le y^-<x-n2^{-n},

1868: $$

1869: Then, $Z^{e^-}_s<x<Z_s^{e^+}$, so

1870: $$

1871: x-\ve\le Z^{e^-}_t\le Z_t^{(s,x)}\le Z_t^{e^+}\le x+\ve.

1872: $$

1873: Hence $\|\Phi_{(s,t]}-\id\|\le\ve$, as required.

1874: \end{proof}

1875: 

1876: Recall that $\Phi^E$ denotes the common restriction of $\Phi^{E,+}$ and $\Phi^{E,-}$ to $D^\circ_E$.

1877: \begin{proposition}

1878: Let $E=\Q^2$. Then $\Phi^E$ is continuous at $z$ for all $z\in C^\circ_E$.

1879: \end{proposition}

1880: \begin{proof}

1881: Consider a sequence $(z_k:k\in\N)$ in $D^\circ_E$ and suppose that $z_k\to z$ in $D_E$, with $z\in C^\circ_E$.

1882: Set $\phi^k=\Phi^E(z_k)$ and $\phi=\Phi^E(z)$. 

1883: It will suffice to show that, for all $n\in\N$, we have $d_C^{(n)}(\phi^k,\phi)\to0$ as $k\to\infty$.

1884: Given $\ve>0$, choose $\ve'>0$ and $\eta>0$ so that $\ve'+2\eta<\ve$.

1885: Then choose $m\in\N$ so that $\ve'+2\eta+1/m<\ve$ and so that $\|\phi_{ts}-\id\|<\eta$ for all $s,t\in(-n,n)$

1886: with $0<t-s<1/m$. Consider the finite set

1887: $$

1888: F=(m^{-1}\Z\cap[-n,n))\times(m^{-1}\Z\cap[0,1)).

1889: $$

1890: There exists a $K<\infty$ such that, for all $k\ge K$, all $(s_0,x_0)\in F$, and all $t\in(s_0,n]$,

1891: $$

1892: |\phi^{k,+}_{(s_0,t]}(x_0)-\phi_{ts}^+(x_0)|=|\phi^{k,-}_{(s_0,t]}(x_0)-\phi_{ts}^-(x_0)|<\ve'.

1893: $$

1894: For all $s\in[-n,n)$ and all $x\in[0,1)$, there exists $(s_0,x_0)\in F$ such that

1895: $$

1896: s_0\le s<s_0+1/m,\q x_0\le x+\ve'+\eta+1/m<x_0+1/m.

1897: $$

1898: Then

1899: $$

1900: \phi^{k,+}_{(s_0,s]}(x_0)\ge\phi^+_{ss_0}(x_0)-\ve'\ge x_0-\ve'-\eta>x,

1901: $$

1902: so

1903: $$

1904: \phi^{k,+}_{(s_0,t]}(x_0)\ge \phi^{k,+}_{(s,t]}(x),\q t\ge s.

1905: $$

1906: Also, we have

1907: $$

1908: \phi_{ss_0}^+(x_0)\le x_0+\eta\le x+\ve'+2\eta+1/m<x+\ve,

1909: $$

1910: so

1911: $$

1912: \phi^+_{ts_0}(x_0)\le\phi^+_{ts}(x+\ve),\q t\ge s.

1913: $$

1914: Now, for all $t\in(s,n]$,

1915: $$

1916: \phi_{(s_0,t]}^{k,+}(x_0)\le\phi_{ts_0}^+(x_0)+\ve,

1917: $$

1918: so

1919: $$

1920: \phi_{(s,t]}^{k,+}(x)\le\phi_{ts}^+(x+\ve)+\ve.

1921: $$

1922: By a similar argument, for all $t\in(s,n]$,

1923: $$

1924: \phi_{(s,t]}^{k,-}(x)\ge\phi_{ts}^-(x-\ve)-\ve,

1925: $$

1926: so $d_\cD(\phi^k_{(s,t]},\phi_{(s,t]})\le\ve$. Hence $d_C^{(n)}(\phi^k,\phi)\to0$ as $k\to\infty$, as required.

1927: \end{proof}

1928: 

1929: 

1930: \subsection{From the coalescing Brownian flow to the Brownian webs}

1931: The coalescing Brownian flow, which provides the limit object for our main result

1932: is a refinement, in certain respects, of Arratia's flow of coalescing Brownian motions,

1933: via the work of T\' oth and Werner. In a series of works, beginning with \cite{FINR}, Fontes, Newman and others

1934: have already provided another such refinement, in fact several, which they call Brownian webs.

1935: They direction taken by Fontes et al. emphasises path properties: 

1936: the Brownian web is conceived as a random element of a space $\cH$ of compact collections of

1937: $\R$-valued paths with specified starting points. 

1938: In our formulation, one does not see so clearly the possible varieties of path, but we find the

1939: state space $C^\circ(\R,\cD)$ a convenient one, with a natural time-reversal map, and just one

1940: candidate for a probability measure corresponding to Arratia's flow.

1941: We now attempt to clarify the relationship between 

1942: the coalescing Brownian flow and one of the Brownian webs.

1943: 

1944: In \cite{FINR}, the authors make the comment that there is more than

1945: one natural distribution of an $\cH$-valued\footnote{We refer to \cite{FINR} for precise definitions and attempt here only to

1946: give a flavour of their results.} random variable $\cW$ that satisfies the following

1947: two conditions.

1948: \begin{enumerate}

1949:   \item[(i)]

1950:    From any deterministic point $(x,t)$ in space-time, there is almost

1951:    surely a unique path $W_{x,t}$ in $\cW$ starting from $(x,t)$.

1952:   \item[(ii)]

1953:    For any deterministic $n$ and $(x_1, t_1), \ldots, (x_n, t_n)$, the

1954:    joint distribution of $W_{x_1, t_1}, \ldots, W_{x_n, t_n}$ is that

1955:    of coalescing Brownian motions (with unit diffusion constant).

1956: \end{enumerate}

1957: The {\em standard Brownian web} satisfies (i) and (ii) together with a certain

1958: minimality condition.

1959: On the other hand, the {\em forward full Brownian web}, introduced in \cite{FN},

1960: satisfies (i) and (ii) together with a certain maximality condition, subject to a

1961: non-crossing condition.

1962: Also in \cite{FN}, the authors characterize a third object, the {\em full

1963: Brownian web}, which is a random variable on the space $\cH^F$ of

1964: compact collections of paths from $\R\to\R$, and explain how this is naturally related

1965: to the other Brownian webs.

1966: 

1967: We now describe in detail the space $\cH^F$ of the full Brownian web and give its characterization. 

1968: We then show that there is a natural way to realise the full Brownian web (and hence also the standard 

1969: Brownian web and forward full Brownian web) as a random variable on $C^\circ(\R,\cD)$. 

1970: Define the function $\Phi : [-\infty, \infty]\times\R \rightarrow [0,1]$ by

1971: $$

1972: \Phi(x,t)=\frac{\tanh(x)}{1+|t|}.

1973: $$

1974: Now construct the two metric spaces $(\Pi^F,d)$ and $(\cH^F,d_{\cH^F})$ as follows. 

1975: Let $\Pi^F$ denote the set of functions $f:\R\rightarrow[-\infty,\infty]$ such that $\Phi(f(t),t)$

1976: is continuous, and define

1977: $$

1978: d^F(f_1,f_2)= \sup_{t\in\R}|\Phi(f_1(t),t)-\Phi(f_2(t),t)|.

1979: $$

1980: Then $(\Pi^F,d^F)$ is a complete separable

1981: metric space.

1982: Now let $\cH^F$ denote the set of compact

1983: subsets of $(\Pi^F,d^F)$, with $d_{\cH^F}$ the induced Hausdorff metric

1984: $$

1985: d_{\cH^F}(K_1,K_2)=\sup_{g_1\in K_1}\inf_{g_2\in K_2}d^F(g_1,g_2)\vee

1986:               \sup_{g_2\in K_2}\inf_{g_1\in K_1}d^F(g_1,g_2).

1987: $$

1988: The space $(\cH^F,d_{\cH^F})$ is also complete and separable. Let $\cF_{\cH^F}$ be the Borel $\sigma$-algebra on $\cH^F$.

1989: The full Brownian web is defined in \cite{FN} as follows.

1990: \begin{defn}

1991: A full Brownian web $\bar{\cW}^F$ is any $({\cH^F},{\cF}_{{\cH^F}})$-valued random variable

1992: whose distribution has the following properties.

1993: \begin{itemize}

1994: \item[(a)] Almost surely the paths of $\bar{\cW}^F$ are noncrossing (although

1995: they may touch, including coalescing and bifurcating). 

1996: \item[($b_1$)]

1997: From any deterministic point $(x,t) \in \mathbb{R}^2$,

1998: there is almost surely a unique path $W_{x,t}^F$ passing through

1999: $x$ at time $t$.

2000: \item[($b_2$)]

2001: For any deterministic $m$, $\{ (x_1,t_1), \ldots, (x_m,t_m) \}$,

2002: the joint distribution of the semipaths $\{ W^F_{x_j, t_j}(t), t

2003: \geq t_j, j=1, \dots, m \}$ is that of a flow of coalescing Brownian

2004: motions (with unit diffusion constant).

2005: \end{itemize}

2006: \end{defn}

2007: They show that any two full Brownian webs have the same distribution. We make a variation of this definition, to tie up with our

2008: work, in requiring in ($b_2$) that the semipaths have the law of coalescing Brownian motions {\em on the circle}.

2009: 

2010: Fix $E$, a countable subset of $\R^2$ containing $\Q^2$. 

2011: We define a map $\theta_{F} : C^\circ(\R,\cD) \rightarrow \cH^F$ as follows.

2012: Let

2013: $$

2014: \Pi = \bigcup_{t \in \R} C([t, \infty), [-\infty, \infty]).

2015: $$

2016: Let $\bar{\cH}$ be the set of all subsets $A$ of $\Pi\cup\Pi^F$ having the the following

2017: noncrossing property

2018: $$

2019: \text{ for all $f,g\in A$ and all $s,t\in\dom(f)\cap\dom(g)$, $f(s)<g(s)$ implies $f(t)\le g(t)$.}

2020: $$

2021: We say that a set $U \in \bar{\cH}$ is {\em maximal} if, 

2022: for any $f\in \Pi\cup\Pi^F$, $U \cup \{ f \} \in \bar{\cH}$ implies $f \in U$.

2023: For $\phi \in C^\circ(\R,\cD)$, define $\th_0(\phi)=\{Z^{e,+}(\phi):e\in E\}$.  

2024: Note that, since $E$ is dense in $\R^2$, if $f,g\in\Pi\cup\Pi^F$ and $\th_0(\phi)\cup\{f\},\th_0(\phi)\cup\{g\}\in\bar{\cH}$,

2025: then also $\th_0(\phi)\cup\{f,g\}\in\bar{\cH}$.

2026: Hence there exists a unique maximal set $\theta_{FF}(\phi)\in\bar{\cH}$ containing $\th_0(\phi)$. Furthermore, this set is independent of the choice of $E$.

2027: Define $\theta_F(\phi) = \theta_{FF}(\phi) \cap \Pi^F$.

2028: 

2029: This can be viewed as a random variable on $C^\circ(\R,\cD)$ as a consequence of Propositions \ref{thetainh} and \ref{isomet} below. 

2030: We note that, under the measure $\mu_W$, the conditions to be a full Brownian web are almost surely satisfied: 

2031: (a) is immediate from the construction; 

2032: ($b_1$) follows from Proposition \ref{FSX}, where condition (i) holds almost surely by Proposition \ref{MPI}, and condition (ii) depends on a countable number of almost sure conditions; 

2033: ($b_2$) follows from Theorem \ref{UBPR}. In fact $\theta_{FF}$ defined above is a forward full Brownian web, 

2034: and the standard Brownian web can be constructed similarly, but we omit the details here.

2035: 

2036: \begin{prop}

2037: \label{thetainh}

2038: The set $\theta_{F}(\phi) \subset \Pi^F$ is compact.

2039: \end{prop}

2040:   \begin{proof}

2041:    Suppose that $f_1, f_2, \ldots \in \theta_{F}(\phi)$. Since paths in

2042:    $\theta_{F}(\phi)$ are noncrossing, there exists a subsequence $n_r$

2043:    such that $f_{n_r}(t)$ is monotone for all $t \in \R$ and so

2044:    there exists some $f:\R \rightarrow [-\infty, \infty]$ such that $f_{n_r}

2045:    \rightarrow f$ pointwise.

2046: 

2047:    Since $\phi_{ts}(\cdot)$ is periodic with period 1, if $g \in \theta_{F}(\phi)$, then $g+m \in \theta_{F}(\phi)$ for all $m \in \mathbb{Z}$. Suppose $f(s) = \infty$ for some $s \in \mathbb{R}$. By restricting to a further subsequence if necessary, we may assume $f_{n_r}(s) \geq f_{n_{r-1}}(s) + 1$ for all $r$. But then, by noncrossing, $f_{n_r}(t) \geq f_{n_{r-1}}(t) + 1$ for all $t \in \mathbb{R}$. Therefore $f(t) = \infty$ for all $t \in \mathbb{R}$ and $f_{n_r} \rightarrow f$ in $(\Pi^F,d^F)$. A similar argument applies if $f(s)=-\infty$ for some $s \in \mathbb{R}$. Hence we may assume $|f(t)| < \infty$ for all $t \in \mathbb{R}$.

2048: 

2049:    Note that $d^F(f,g) \leq d^{(n)}(f,g) \vee 2/(n+1)$ where

2050:    $$

2051:    d^{(n)}(f,g) = \sup_{-n < t < n} |f(t) - g(t)|.

2052:    $$

2053:    Given $\epsilon > 0$, there exist $n > 2\epsilon^{-1}$, and $-n = a_0 < \cdots < a_M=n$ such that

2054:    $\| \phi^+_{s a_k} - \id \| < \frac{\epsilon}{4}$ for all $a_k \leq s \leq

2055:    a_{k+1}$, $k = 0, \ldots, M-1$. Pick $N$ sufficiently large

2056:    that if $n_r > N$, $|f_{n_r}(a_k)-f(a_k)| <

2057:    \frac{\epsilon}{4}$ for $k = 1, \ldots, M-1$. Then for $a_k

2058:    \leq s < a_{k+1}$,

2059:    $$

2060:    |f_{n_r}(s) - f(s)| \leq |\phi^+_{s a_{k}}(f(a_k) +

2061:    \tfrac{\epsilon}{4}) - \phi^+_{s a_{k}}(f(a_k) - \tfrac{\epsilon}{4})|

2062:    < \epsilon.

2063:    $$

2064:    Hence, $f_{n_r} \rightarrow f$ in $(\Pi^F,d^F)$. Therefore, $f$

2065:    is continuous and $\theta_{FF}(\phi) \cup \{ f \}$ is noncrossing,

2066:    and so $f \in \theta_F (\phi)$ proving compactness.

2067: \end{proof}

2068: 

2069: \begin{prop}

2070: \label{FSX}

2071: For every $(x,s) \in \mathbb{R}^2$, there exists some $f \in \theta_F (\phi)$ with $f(s)=x$. Moreover, $f$ is unique if the following two conditions hold.

2072: \begin{enumerate}

2073:  \item[(i)] $\phi^+_{ts}(x) = \phi^+_{ts}(x)$ for all $t \geq s$;

2074:  \item[(ii)] $\phi^+_{st}(y) \neq x$ for all $(y,t) \in \mathbb{Q}^2$ with $t < s$.

2075: \end{enumerate}

2076: \end{prop}

2077: \begin{proof}

2078:  Define a function $f:\mathbb{R} \rightarrow \mathbb{R}$ by

2079:  $$

2080:  f(t) = \begin{cases}

2081:          \inf \{ y : \phi_{st}^+(y) > x \} &\text{if $t < s$} \\

2082:          \phi^+_{ts}(x) &\text{if $t \geq s$}.

2083:         \end{cases}

2084:  $$

2085: We first show that $f$ is continuous. Clearly this is the case on $[s, \infty)$. Suppose $t < u \leq s$. Pick a sequence $y_n \downarrow f(t)$. By the weak flow property of $\phi$, $\phi^+_{su}(\phi^+_{ut}(y_n)) \geq \phi^+_{st}(y_n)>x$ and so $\phi^+_{ut}(y_n) \geq f(u)$ for all $n$. Letting $n \rightarrow \infty$ gives $\phi^+_{ut}(f(t)) \geq f(u)$. Similarly, using $\inf \{ y : \phi_{st}^+(y) > x \}=\sup \{ y : \phi_{st}^-(y) \leq x \}$, $\phi^-_{ut}(f(t)) \leq f(u)$. Now given $\epsilon > 0$, there exists some $\delta > 0$ such that $\| \phi_{ut} - \id \| < \epsilon$ for all $0 < u -t < \delta$. But then $|f(t) - f(u)| \leq |f(t) - \phi^-_{ut}(f(t))|\wedge|f(t) - \phi^+_{ut}(f(t))|< \epsilon$, proving continuity. 

2086: Also, by construction, $\th_0(\phi)\cup\{f\}\in\bar{\cH}$ and so $f \in \theta_F (\phi)$ as required.

2087: 

2088: For uniqueness, suppose $f_1, f_2 \in \theta_F (\phi)$ with $f_1(s)=f_2(s)=x$. By the noncrossing property, for all $t \geq s$, $\phi_{ts}^-(x) \leq f_i(t) \leq \phi_{ts}^+(x)$. Condition (i) therefore implies $f_1(t)=f_2(t)$ for all $t \geq s$. If $t < s$, then $\sup \{ y : \phi_{st}^-(y) < x \} \leq f_i(t) \leq \inf \{ y : \phi_{st}^+(y) > x \}$. If $\inf \{ y : \phi_{st}^+(y) > x \} - \sup \{ y : \phi_{st}^-(y) < x \} > 0$, then by continuity, there exist rationals $(y,u)$ with $\phi^+_{st}(y) = x$. Hence (ii) ensures $f_1(t)=f_2(t)$ for all $t < s$.

2089: \end{proof}

2090: 

2091: 

2092:   \begin{prop}

2093:   \label{isomet}

2094:    The function $\theta_{F} : C^\circ(\R,\cD) \rightarrow \cH^F$ is continuous.

2095:   \end{prop}

2096:   \begin{proof}

2097:    A sequence $\phi^m \rightarrow \phi$ in $(C^\circ(\R,\cD), d_C)$ if and only if $d^{(n)}_C(\phi^m, \phi) \rightarrow 0$ for all $n \in \mathbb{N}$. Since $d^F(f,g) \leq d^{(n)}(f,g) \vee 2/(n+1)$, where $d^{(n)}$ is defined in the previous proposition, in order to show that $\theta_{F}$ is continuous, it is enough to show that for every $n$, $d_{\cH^F}^{(n)}(\theta_{F}(\phi^1),\theta_{F}(\phi^2)) \leq d^{(n)}_C(\phi^1, \phi^2)$ for all $\phi^1, \phi^2 \in C^\circ(\R,\cD)$, where

2098:    $$

2099:    d^{(n)}_{\cH^F}(K_1,K_2)=\sup_{g_1\in K_1}\inf_{g_2\in K_2}d^{(n)}(g_1,g_2)\vee

2100:               \sup_{g_2\in K_2}\inf_{g_1\in K_1}d^{(n)}(g_1,g_2).

2101:    $$

2102:    

2103: Suppose $\phi^1, \phi^2 \in C^\circ(\R,\cD)$. By compactness, there exist $f_1 \in \theta_{F}(\phi^1)$, $f_2 \in \theta_{F}(\phi^2)$ such that $d^{(n)}(f_1,f_2) = d^{(n)}_{\cH^F}(\theta_{F}(\phi^1), \theta_{F}(\phi^2))$. Without loss of generality, suppose that $f_2$ is

2104: chosen so that $d^{(n)}(f_1,f_2) = \inf_{g \in \theta_{F}(\phi^2)} d^{(n)}(f_1,g) = f_2(s) - f_1(s)$ for some $s \in \mathbb{R}$. Let $f_1(s)=x$ and $f_2(s)=y$. By continuity, there exists some $t > s$ such that $f_1(t) - \phi^-_{ts}(y) = d^{(n)}(f_1,f_2)$, otherwise it would be possible to pick some $g \in \theta_{F}(\phi^2)$ with $d^{(n)}(f_1,g) < d^{(n)}(f_1,f_2)$.

2105: Since $f_1(t) \in [\phi^{1,-}_{ts}(x), \phi^{1,+}_{ts}(x)]$,

2106:    $$

2107:    \phi^{2,-}_{ts}(y) - \phi^{1,+}_{ts}(x) \leq x-y \leq \phi^{2,+}_{ts}(y) -

2108:    \phi^{1,-}_{ts}(x).

2109:    $$

2110:    Hence there exists some

2111:    $$

2112:    u \in \left [ \frac{1}{2}(x + \phi^{1,-}_{ts}(x)), \frac{1}{2}(x +

2113:    \phi^{1,+}_{ts}(x)) \right ] \cap \left [ \frac{1}{2}(y +

2114:    \phi^{2,-}_{ts}(y)), \frac{1}{2}(y + \phi^{2,+}_{ts}(y)) \right ].

2115:    $$

2116:    Therefore,

2117:    \begin{eqnarray*}

2118:      d_{\cH^F}(\theta_F(\phi^1), \theta_F(\phi^2))

2119:        &=& |x-y| \\

2120:        &=& |\phi^{1\times}_{ts}(u) -\phi^{2\times}_{ts}(u)| \\

2121:     &\leq& \| \phi^{1\times}_{ts} - \phi^{2\times}_{ts} \| \\

2122:     &\leq& d^{(n)}_{C}(\phi^1, \phi^2),

2123:    \end{eqnarray*}

2124:    as required.

2125:   \end{proof}

2126: 

2127: 

2128: \markright{\sc{Bibliography}}

2129: 

2130: 

2131: \def\cprime{$'$}

2132: \begin{thebibliography}{10}

2133: 

2134: \bibitem{A79}

2135: Richard~A. Arratia.

2136: \newblock {\em Coalescing {B}rownian motions on the line}.

2137: \newblock PhD thesis, University of Wisconsin, Madison, 1979.

2138: 

2139: \bibitem{B+C}

2140: Martin~Z. Bazant and Darren Crowdy.

2141: \newblock Conformal mapping methods for interfacial dynamics.

2142: \newblock \texttt{arXiv:cond-mat/0409439}, 2005.

2143: 

2144: \bibitem{B}

2145: Patrick Billingsley.

2146: \newblock {\em Convergence of probability measures}.

2147: \newblock Wiley Series in Probability and Statistics: Probability and

2148:   Statistics. John Wiley \& Sons Inc., New York, 1999.

2149: 

2150: \bibitem{C+M}

2151: L.~Carleson and N.~Makarov.

2152: \newblock Aggregation in the plane and {L}oewner's equation.

2153: \newblock {\em Comm. Math. Phys.}, 216(3):583--607, 2001.

2154: 

2155: \bibitem{Eden}

2156: Murray Eden.

2157: \newblock A two-dimensional growth process.

2158: \newblock In {\em Proc. 4th Berkeley Sympos. Math. Statist. and Prob., Vol.

2159:   IV}, pages 223--239. Univ. California Press, Berkeley, Calif., 1961.

2160: 

2161: \bibitem{FINR}

2162: L.~R.~G. Fontes, M.~Isopi, C.~M. Newman, and K.~Ravishankar.

2163: \newblock The {B}rownian web: characterization and convergence.

2164: \newblock {\em Ann. Probab.}, 32(4):2857--2883, 2004.

2165: 

2166: \bibitem{FN}

2167: Luiz~Renato Fontes and Charles~M. Newman.

2168: \newblock The full {B}rownian web as scaling limit of stochastic flows.

2169: \newblock {\em Stoch. Dyn.}, 6(2):213--228, 2006.

2170: 

2171: \bibitem{harris}

2172: Theodore~E. Harris.

2173: \newblock Coalescing and noncoalescing stochastic flows in {${\bf R}\sp{1}$}.

2174: \newblock {\em Stochastic Process. Appl.}, 17(2):187--210, 1984.

2175: 

2176: \bibitem{HL}

2177: M.~B. Hastings and L.~S. Levitov.

2178: \newblock {L}aplacian growth as one-dimensional turbulence.

2179: \newblock {\em Physica D}, 116 (1-2):244, 1998.

2180: 

2181: \bibitem{MR1943877}

2182: Jean Jacod and Albert~N. Shiryaev.

2183: \newblock {\em Limit theorems for stochastic processes}, volume 288 of {\em

2184:   Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of

2185:   Mathematical Sciences]}.

2186: \newblock Springer-Verlag, Berlin, 2003.

2187: 

2188: \bibitem{Kesten}

2189: Harry Kesten.

2190: \newblock Hitting probabilities of random walks on {${\bf Z}\sp d$}.

2191: \newblock {\em Stochastic Process. Appl.}, 25(2):165--184, 1987.

2192: 

2193: \bibitem{L}

2194: Gregory~F. Lawler.

2195: \newblock {\em Conformally invariant processes in the plane}, volume 114 of

2196:   {\em Mathematical Surveys and Monographs}.

2197: \newblock American Mathematical Society, Providence, RI, 2005.

2198: 

2199: \bibitem{NPW}

2200: L.~Niemeyer, L.~Pietronero, and H.~J. Wiesmann.

2201: \newblock Fractal dimension of dielectric breakdown.

2202: \newblock {\em Phys. Rev. Lett.}, 52:1033--1036, 1984.

2203: 

2204: \bibitem{Piterbarg}

2205: Vladimir~V. Piterbarg.

2206: \newblock Expansions and contractions of isotropic stochastic flows of

2207:   homeomorphisms.

2208: \newblock {\em Ann. Probab.}, 26(2):479--499, 1998.

2209: 

2210: \bibitem{RZ}

2211: Steffen Rohde and Michel Zinsmeister.

2212: \newblock Some remarks on {L}aplacian growth.

2213: \newblock {\em Topology Appl.}, 152(1-2):26--43, 2005.

2214: 

2215: \bibitem{TW}

2216: B{\'a}lint T{\'o}th and Wendelin Werner.

2217: \newblock The true self-repelling motion.

2218: \newblock {\em Probab. Theory Related Fields}, 111(3):375--452, 1998.

2219: 

2220: \bibitem{Tsirelson}

2221: Boris Tsirelson.

2222: \newblock Nonclassical stochastic flows and continuous products.

2223: \newblock {\em Probab. Surv.}, 1:173--298 (electronic), 2004.

2224: 

2225: \bibitem{W+S}

2226: T.~A. Witten and L.~M. Sander.

2227: \newblock Diffusion-limited aggregation, a kinetic critical phenomenon.

2228: \newblock {\em Phys. Rev. Lett.}, 47(19):1400--1403, 1981.

2229: 

2230: \end{thebibliography}

2231: \end{document}

2232: 

2233: