Brownian Motion in wedges, last passage time and the second arc-sine law
Extracted source text ยท arXiv cond-mat/0302269
\documentclass[11pt]{article}
\usepackage[english,francais]{babel}
\usepackage[latin1]{inputenc}
\usepackage{epsfig,graphicx}
\usepackage{latexsym}
%%%%%%%%%%%%%%%%%%%%%%%%%% Format de la page %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\textwidth= 16cm
\oddsidemargin= 0.5cm
\evensidemargin=-0.5cm
\topmargin=-1cm
\textheight= 24cm
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Fonts %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Fonte de l'AMS
\newfont{\ensmathquatorze}{msbm10 scaled 1400}
\newfont{\ensmathonze}{msbm10 scaled 1100}
\newfont{\ensmathdix}{msbm10}
\newfont{\ensmathneuf}{msbm10 scaled 833}
\newfont{\ensmathhuit}{msbm10 scaled 694}
\newfam\ensmathfam
\textfont\ensmathfam=\ensmathonze
\scriptfont\ensmathfam=\ensmathdix
\scriptscriptfont\ensmathfam=\ensmathhuit
\def\ensmf{\fam\ensmathfam\ensmathonze} % Usage: {\ensmf texte...}
%%%%%%%%%%%%%%%%%%%%%%%%%%% Newcommand, Def %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def\be{\begin{equation}}
\def\ee{\end{equation}}
\def\bea{\begin{eqnarray}}
\def\eea{\end{eqnarray}}
\newcommand{\tr}[1]{\mathop{\mathrm{Tr}}\nolimits\left\{ #1 \right\}} % Trace
\newcommand\lab{l_{\alpha\beta}}
\newcommand\xab{x_{\alpha\beta}}
\newcommand\xba{x_{\beta\alpha}}
\newcommand\xaa{x_{ab}}
\newcommand\xbb{x_{ba}}
\newcommand\Gab{G_{(\alpha\beta )}}
\newcommand\Gba{G_{(\beta\alpha )}}
\newcommand\Gaa{G_{(ab )}}
\newcommand\Gbb{G_{(ba )}}
\newcommand\Gpaa{G'_{(ab )}}
\newcommand\Gpbb{G'_{(ba )}}
\newcommand\Gpab{G'_{(\alpha\beta )}}
\newcommand\Gpba{G'_{(\beta\alpha )}}
\newcommand\dg{\partial_{\gamma}}
\newcommand\pal{\psi_{\alpha\beta}}
\newcommand\pbe{\psi_{\beta\alpha}}
\newcommand\ppal{\psi'_{\alpha\beta}}
\newcommand\ppbe{\psi'_{\beta\alpha}}
\newcommand\fal{\phi_{(\alpha\beta )}}
\newcommand\fbe{\phi_{(\beta\alpha )}}
\newcommand\fpal{\phi'_{(\alpha\beta )}}
\newcommand\fpbe{\phi'_{(\beta\alpha )}}
\newcommand\cab{\chi_{\alpha\beta}}
\newcommand\cba{\chi_{\beta\alpha}}
\newcommand\cpab{\chi'_{\alpha\beta}}
\newcommand\cpba{\chi'_{\beta\alpha}}
\newcommand\caa{\chi_{ab}}
\newcommand\cbb{\chi_{ba}}
\newcommand\cabi{\chi_{\alpha\beta_i}}
\newcommand\cbia{\chi_{\beta_i\alpha}}
\newcommand\cpabi{\chi'_{\alpha\beta_i}}
\newcommand\cpbia{\chi'_{\beta_i\alpha}}
\newcommand\sg{\sqrt {\gamma }}
\newcommand\pab{\psi_{ab}}
\newcommand\pba{\psi_{ba}}
\newcommand\ppab{\psi'_{ab}}
\newcommand\ppba{\psi'_{ba}}
\newcommand\dpl{ \frac{ \d \psi_{\alpha\beta}}{ \d x_{\alpha\beta}} }
\newcommand\dpt{ \frac{ \d \psi_{\beta\alpha}}{ \d x_{\alpha\beta}} }
\newcommand\dpa{ \frac{ \d \psi_{ab}} { \d x_{ab}} }
\newcommand\dpb{ \frac{ \d \psi_{ba}} { \d x_{ab}} }
\newcommand\dpi{ \frac{ \d \psi_{\alpha\beta_i}}{ \d x_{\alpha\beta_i}} }
\newcommand\dpj{ \frac{ \d \psi_{\beta_i\alpha}}{ \d x_{\alpha\beta_i}} }
\newcommand\wal{W_{\alpha\beta}}
\newcommand\wla{W_{\beta\alpha}}
\newcommand\wab{W_{ab}}
\newcommand\wba{W_{ba}}
\newcommand\Wal{{\cal W}_{\alpha\beta}}
\newcommand\Wab{{\cal W}_{ab}}
\newcommand\Wla{{\cal W}_{\beta\alpha}}
\newcommand\Wba{{\cal W}_{ba}}
\def\d{{\rm d}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
\selectlanguage{english}
\title{ Brownian Motion in wedges, last passage time and the
second arc-sine law }
\author{ Alain Comtet$^{1,2}$ and Jean Desbois $^1$}
\maketitle
{\small
\noindent
$^1$ Laboratoire de Physique Th\'eorique et Mod\`eles Statistiques.
Universit\'e Paris-Sud, B\^at. 100, F-91405 Orsay Cedex, France.
}
{\small
\noindent
$^2$ Institut Henri Poincar\'e, 11 rue Pierre et Marie Curie, 75005, Paris
}
\vskip1.5cm
\begin{abstract}
We consider a planar Brownian motion starting from $O$ at time
$t=0$ and stopped
at $t=1$ and a set $F= \{ OI_i \; ; \ i=1,2,\ldots ,n\}$ of $n$ semi-infinite
straight lines emanating from $O$. Denoting by $g$ the last time when
$F$ is reached by the Brownian motion, we compute the probability
law of $g$. In particular, we show that, for a symmetric $F$ and
even $n$ values, this law can be expressed as a sum of \ $\arcsin $ \
or \ $(\arcsin )^2 $ \ functions. The
original result of Levy is recovered as the
special case $n=2$. A relation with the problem of reaction-diffusion of a
set of three particles in one dimension is discussed.
\end{abstract}
\vskip1.5cm
The first arc-sine law gives the distribution of the number of positive
partial sums in a sequence of independent and identically distributed
random variables. It was first discovered by P. Levy in his study of the
linear Brownian motion and then discussed a lot for its relevance to the
coin-tossing game \cite{feller}. The second arc-sine law, also discovered by
P. Levy \cite{levy},
provides an information on the last passage time which can be stated as
follows. Consider a linear Brownian motion $B(\tau)$ starting at $0$ at
time $\tau=0$ and stopped at time $t$ and let $g$ be the last time when
$0$ is visited.
The random variable
\be\label{a1}
g= \sup \{\ {\tau <t,B(\tau)=0}\}
\ee
satisfies
\bea
P(g<u) &=& \frac{2}{\pi } \arcsin \sqrt{ \frac{u}{t}}
\label{asl} \\
\mbox{with the density } \quad {\cal P}(u) &=& \frac{1}{\pi }
\frac{1}{ \sqrt{u(t-u)} } \label{asl1}
\eea
\vskip.5cm
\noindent
Over the years, this result has been extended in several different directions
(see for instance, \cite{yor,Y2}) and is still a subject of active research in
probability \cite{Y1}.
Generalizations of the first arc-sine law have also
been considered in different contexts
(one-dimensional diffusion in a random medium
\cite{MC}, Brownian motion on graphs \cite{JD} and, also,
in two dimensions \cite{bingham}).
\vskip.5cm
%A physical motivation for this problem comes from the fact that it can be
%mapped onto a problem of diffusion on a line of a set of three independant
%Brownian particles. The last intersection time is the same as the last
%passage time through a suitable cone. This problem has been studied in the
%context of reaction diffusion models. In the case of particles which diffuse
%and annihilate, the
%quantity of main interest is the first
%passage time. Its
%large time behaviour decays with a power law which can be computed exactly.
\noindent
The purpose of this Letter is to present a two dimensional generalization of
the law
(\ref{asl}). As a by-product of this result we also derive an explicit
expression of the
first passage time distribution which is relevant for a problem of
reaction-diffusion involving three identical particles.
\noindent
Exit problems for Brownian motion have a rich history and several
applications in physics (see for instance \cite{RE}). They are in particular
related with problems of
capture of independent Brownian particles diffusing on the line. This
connection was first anticipated by Arratia \cite {A} and then
discussed in the
mathematics \cite {NOC,B} and physics litterature \cite {RE,F,BA,M}
mainly in the context of reaction-diffusion
models. In the case of three particles, the process
$(x_1(t)-x_2(t)$,
$x_2(t)-x_3(t))$ defines a certain diffusion in a quadrant of $R^2$.
By a suitable transformation, this process can be mapped on a diffusion
inside a wedge whose angle depends on the diffusion constants. Using this
correspondance, it has been shown that the first passage time through the
wedge gives the survival probability; a quantity which decays with a power
law which only depends on the angle of the wedge \cite {BA,DB}.
At the end of this work, we exploit this correspondance to compute exactly
(and not only asymptotically)
the first collision time distribution for a three particle problem. Our
approach is based on an identity relating the first passage and last passage
distribution which has an interesting probabilistic interpretation\cite {Y3}.
\vskip.5cm
To begin with, let us start by considering,
as in Figure $1$, a wedge of apex $O$ and angle $\phi$
with a boundary $F= \{ OI_i \; ; \ i=1,2 \} $ and a two dimensional
Brownian motion
$\overrightarrow{r}(t) $
starting from $O$ at $t=0$ and stopped at $t$ somewhere in the plane.
We denote by $g$ the last time when $F$ is visited and compute
the probability $P(g<u)$. Due to the scaling property of the
Brownian motion, this distribution is a function of the reduced variable
$u/t$. In the following, we will for simplicity set $t=1$.
\begin{figure}[!h]
\begin{center}
\includegraphics[scale=.4,angle=0]{sector1.pstex}
\caption{ The frontier $F= \{ OI_1, OI_2\}$ of the wedge divides the plane
into 2 regions, {\bf (1)} and {\bf (2)}. }
\end{center}
\end{figure}
\vskip.5cm
\noindent
Suppose that the particle reaches some point $\overrightarrow{r_0}$ at time
$t=u$ (see Figure 1).
Clearly, if $\overrightarrow{r_0}$ belongs to region {\bf (1)}
(resp.{\bf (2)}), the particle must stay in {\bf (1)}
(resp.{\bf (2)}) between $t=u$ and $t=1$ in order to satisfy the
condition $g<u$. We can therefore write
\be\label{n1}
P(g<u) = P^{(1)}(u) + P^{(2)}(u)
\ee
\vskip.3cm
\noindent
Expressing the fact that the propagation is free between $t=0$ and $t=u$
and that the particle has not hit the boundary between $t=u$ and $t=1$, we get
\be\label{n2}
P^{(i)}(u)=\int_{(i)} \d ^2 \overrightarrow{r_0}
\int_{(i)} \d ^2 \overrightarrow{r} \; \frac{1}{2\pi u}
\; e^{- \frac{r_0^2}{2u}} \; G^{(i)}( \overrightarrow{r},1;
\overrightarrow{r_0} ,u) \qquad ; \quad i=1,2
\ee
\vskip.2cm
\noindent
The propagator $G^{(1)}$ satisfying the diffusion equation
with Dirichlet boundary conditions on $F$ is given by
\be\label{n3}
G^{(1)} = \frac{2}{\phi \; (1-u)} \; \sum_{m=1}^{\infty }
\sin \frac{m\theta\pi }{\phi } \; \sin \frac{m\theta_0\pi }{\phi }
\; e^{-\frac{r^2+r_0^2}{2(1-u)}} \; I_{\frac{m\pi }{\phi }} \left(
\frac{rr_0}{1-u} \right)
\ee
where $I_{\nu }$ is a modified Bessel function and the notations are
defined on Figure 1.
\vskip.8cm
\noindent
Performing the spatial integrations in (\ref{n2}), we get \cite{GRAD}
\be\label{n4}
P^{(1)}(u) = \frac{1}{\pi \; \phi} \; \sum_{p,k=0}^{\infty }
u^{p\frac{\pi}{\phi} + k + \frac{\pi}{2 \phi} }
\; \frac{ \left[ \Gamma (p\frac{\pi}{\phi} + k + \frac{\pi}{2 \phi} )
\right]^2 }
{\Gamma (\frac{2 p \pi}{\phi} + \frac{\pi}{ \phi} +k+1 ) }
\; \frac{1}{k!}
\ee
$P^{(2)} $ is obtained by the change $\phi \to 2\pi - \phi$ in
(\ref{n4}). Therefore, for arbitrary values of $\phi $
the law $P(g<u) $ is written in terms of a double series.
\vskip.3cm
\noindent
As a check, let us first consider the special case $\phi= \pi$. We may
write
\bea
P(g<u) = 2 P^{(1)}(u) &=& \frac{2}{\pi^2} \; u^{1/2} \sum_{p,k=0}^{\infty }
\frac{u^{p+k}}{k!} \frac{ \left[ \Gamma (p+k+1/2) \right]^2 }
{ (2p+k+1)!} \label{n5} \\
&=& \frac{2}{\pi } \arcsin \sqrt{u} \label{n6}
\eea
\noindent
The fact that one recovers Levy's second arc-sine law is not surprising since, when
$\phi= \pi$, $F$ divides the plane into two half-planes. Therefore the
component
of the Brownian motion parallel to $F$ factorises and plays no role: we are thus left
with a one dimensional problem.
\vskip.5cm
\noindent
Coming back to general values of $\phi $, we can derive the behavior of
the probability density ${\cal P}$ $ ( \equiv \frac{\d P(g<u)}{\d u} ) $
when $u\to 0$ and $u\to 1$. By using (\ref{n1}) and (\ref{n4}), one gets
a power-law behavior when $u\to 0^+$
\bea
{\cal P}(u) &\sim& \frac{1}{\pi } \; u^{-1/2}
\qquad \mbox{for } \ \phi = \pi
\label{n7} \\
{\cal P}(u) &\sim& C(\mu ) \; u^{ \frac{\mu }{2} -1}
\qquad \mbox{for } \ \phi \ne \pi
\label{n8} \\
\mbox{with } \quad C(\mu )
&=& \frac{\mu^2}{2\pi^2} \;
\frac{ \left[ \Gamma(\frac{\mu }{2}) \right]^2 }{\Gamma (\mu +1)}
\label{n80} \\
\mbox{and } \quad \mu &=& \frac{\pi }{ 2\pi - \phi} \qquad \mbox{when}
\quad 0< \phi < \pi
\label{n9} \\
\mu &=& \frac{\pi }{ \phi} \qquad \mbox{when}
\quad \pi < \phi < 2 \pi
\label{n10}
\eea
\vskip.5cm
\noindent
Now, for the limit $u \to 1^- $, using asymptotic expansions for $\Gamma $
functions and also an equivalence between series and integrals, we get
\be\label{n11}
{\cal P}(u) \sim \frac{1 }{ \pi} \; \frac{1 }{ \sqrt{1-u}}
\ee
\vskip.2cm
\noindent
i.e. the same behavior as for (\ref{asl1}). The expression (\ref{n11})
doesn't depend on $\phi$ and we have already seen that $\phi =\pi$ gives
the Levy's law. Remark that $u \to 1^- $ corresponds to Brownian curves
that stop close to $F$.
Therefore, between $t=u$ and $t=1$, the Brownian particle
only ``sees'' an infinitesimal part of $F$, i.e. a straight line as for
$\phi =\pi$. This is, in our opinion, why the result (\ref{n11})
doesn't depend on $\phi$. Actually, it only depends on the fact that the
plane is divided by $F$ into two regions. We will come back to this point
latter on.
\vskip.5cm
\noindent
\begin{figure}[!h]
\begin{center}
\includegraphics[scale=.4,angle=0]{sector2.pstex}
\caption{ $F$ consists in $n$ semi-infinite straight lines starting from $O$. }
\end{center}
\end{figure}
\vskip.5cm
\noindent
To go further, let us remark that for $F= \{ OI_i \; ; \ i=1,2,\ldots ,n\}$
as in Figure 2, (\ref{n1}) becomes simply:
\be\label{r1}
P(g<u) = \sum_{i=1}^n \; P^{(i)}(u)
\ee
\vskip.3cm
\noindent
(Replace $\phi $ by $\phi_i $ in (\ref{n4}) in order to get $P^{(i)}$).
\vskip.7cm
\noindent
Let us now specialize to the situation when $F$ is symmetric and $n$ is even
($n \equiv 2l$). In that case, $F$ consists in $l$ infinite straight lines
crossing at point $O$ and dividing the plane into $2l$ equal angular
sectors, each one of angle $\phi = \pi /l$.
\begin{figure}[!h]
\begin{center}
\includegraphics[scale=.4,angle=0]{sector3.pstex}
\caption{ $F$ is symmetric. The analytic form of $P(g<u)$
will depend on the parity of $l$. Thus, it will be different for the cases a)
and b). For further explanations, see text.}
\end{center}
\end{figure}
\vskip.3cm
\noindent
Equation (\ref{r1}) writes
\be\label{r2}
P(g<u) \equiv P_l(u) =
\frac{2}{\pi^2} \; l^2 \; u^{l/2} \sum_{p=0}^{\infty } u^{lp}
\sum_{k=0}^{\infty }
\frac{ \left[ \Gamma (lp+k+l/2) \right]^2 }
{ (2lp+l+k)!}
\frac{u^{k}}{k!}
\ee
\vskip.6cm
\noindent
$l$ being an integer, we can sum the series and, finally, get
\bea
P_l (u) &=& \frac{2l }{\pi } \left(
\sum_{k=0}^{l-1} \arcsin \left( \sqrt{u} \cos \frac{2\pi k}{l} \right)
\right) \qquad , \quad l \mbox{ odd} \label{r3} \\
P_l (u) &=& \frac{2l }{\pi ^2} \left(
\sum_{k=0}^{l-1} \; (-1)^k \;
\left( \arcsin \left( \sqrt{u} \cos \frac{\pi k}{l} \right) \right)^2
\right) \qquad , \quad l \mbox{ even} \label{r4}
\eea
which is the central result of this paper.
\vskip.3cm
\noindent
We remark that the correct small $u$ behavior for $P_l(u)$ follows from the two
identities
\bea
\sum_{k=0}^{l-1} \left( \cos \frac{2\pi k}{l} \right)^m &=& 0
\qquad , \quad l \mbox{ odd } \quad , \quad m=1,3,\ldots , l-2 \\
\sum_{k=0}^{l-1} (-1)^k \; \left( \cos \frac{\pi k}{l} \right)^m &=& 0
\qquad , \quad l \mbox{ even } \quad , \quad m=0,2,4,\ldots , l-2
\eea
\vskip.6cm
\noindent
In particular
\bea
P_1 (u) &=& \frac{2}{\pi } \arcsin \sqrt{u} \label{r5} \\
P_2 (u)&=& \frac{4}{\pi ^2 } \left( \arcsin \sqrt{u} \right)^2 =
P_1^2 \label{r6} \\
P_3 (u)&=& \frac{6}{\pi } \left(
\arcsin \sqrt{u} \; - \; 2 \;
\arcsin \frac{\sqrt{u}}{2} \right) \label{r7} \\
P_4 (u)&=& \frac{8}{\pi ^2 }\left(
\left( \arcsin \sqrt{u} \right)^2 \; - \; 2 \;
\left( \arcsin \sqrt{\frac{u}{2}} \right)^2 \right)
\label{r8} \\
P_5 (u)&=&
\frac{10}{\pi }\left(
\arcsin \sqrt{u} \; - \; 2 \; \arcsin \left( \cos \frac{\pi }{5}
\; \sqrt{u} \right)
\; + \; 2 \; \arcsin \left( \cos \frac{2 \pi }{5}
\; \sqrt{u} \right)
\right)
\label{r9}
\eea
\noindent
These functions are displayed below in Figure 4.
\begin{figure}[!h]
\begin{center}
\includegraphics[scale=.5,angle=0]{sector4.pstex}
\caption{ The distribution functions $P_l(u)$ for $l=1,\ldots ,5$. }
\end{center}
\end{figure}
\vskip.3cm
\noindent
As expected, the Levy's second arc-sine law is recovered in (\ref{r5}).
Moreover, the result (\ref{r6}) is straightforward since, when $l=2$, the
$2$ components of the Brownian motion factorize. Thus, for $l=2$,
$(\arcsin )^2$ functions appear. What is surprising is that they will
appear each time $l$ is even while being absent when $l$ is odd.
\vskip.3cm
\noindent
For the probability density, ${\cal P}_l $
($\equiv \frac{\d P_l}{\d u}$), with (\ref{r3}) and (\ref{r4}), we obtain:
\be\label{r10}
{\cal P}_l (u) \quad \sim \quad \frac{l}{\pi} \; \frac{1}{\sqrt{1-u}}
\qquad \mbox{ when } \qquad u \to 1^-
\ee
\vskip.3cm
\noindent
This is consistent with (\ref{n11}) that corresponds to $l=1$.
We now present a formula which relates the first passage and the last
passage time distribution. The starting point is (\ref{n1}) and (\ref{n2})
which may be rewritten as
\be\label{r11}
P(g< u)=\int{ Pr(T> 1-u|{r_0} )
\; \frac{1}{u} \: e^{- \frac{r_0^2}{2u}}\:{r_0} \:\d {r_0}}
\ee
where $Pr(T> (1-u)|{r_0})$ is the probability distribution of the first
passage time $T$ through F, given that the process starts at $r_0$.
Then, by scaling one has
\be\label{r12}
Pr(T> (1-u)|{r_0})= Pr(T> {(1-u)\over{r_0}^2}|1)
\ee
By a simple change of variables it follows that
\be\label{r13}
P(g< {1\over1+t})=\int{ Pr(T> {t \over2x}|1 )
\; e^{-x} \:\d x}
\ee
Therefore
\be\label{r14}
P(g< {1\over1+t})=
E(e^{-{t\over 2T}})
\ee
which is a relation between the first passage characteristic function
for a process starting at $r_0=1$ and the probability distribution of the
last passage time. Interestingly enough this formula can also be derived in
a more intrinsic fashion using only time inversion and scaling
\cite {Y3}.
As an application, let us the derive the density of first passage time in a
wedge of angle $\phi= {\pi\over3 }$.
In the context of the capture problem mentioned in the
introduction, this corresponds to
a set of three identical and independent particles \cite{NOC}.
In this case, the distribution $P(g)$ is given in
eq (\ref{r7}). By an inverse Laplace transform (\ref{r14}) gives the
density of first passage time:
\be\label{r15}
f(T)={ 6\over {\pi^{3\over2} T}}\; e^{-{1\over 2T}}\:(\:\int_{0}^{\sqrt
{1\over2T}} e^{y^2}\d y -2\int_{0}^{\sqrt
{1\over8T}} e^{y^2}\d y\;)
\ee
One can check that this formula is in agreement with (16) of \cite{NOC}
which expresses the first collision time probability for a given set of
initial conditions. By averaging this formula over the angle
and setting $r=1$ one recovers (\ref{r15}).
\vskip.3cm
\noindent
ACKNOWLEDGMENT
\noindent
We are particularly grateful to Prof. Marc Yor for introducing us to the
subject and for pointing out (\ref{r14}).
\begin{thebibliography}{10}
\bibitem{feller}
Feller W 1957
\newblock {\it An Introduction to Probability Theory and its Applications}
(John Wiley and Sons: New York)
\bibitem{levy}
Levy P 1948
\newblock{\it Processus Stochastiques et Mouvement Brownien}
(Editions Jacques Gabay: Paris)
\bibitem{yor}
Barlow M, Pitman J and Yor M 1989
\newblock {\it
S\'em. Probabilit\'es XXIII (Lecture Notes in Maths vol 1372)}
(Berlin: Springer)
\bibitem{Y2}
Yor M 1992
\newblock{\it Some Aspects of Brownian Motion}, Part 1, Lectures in
Mathematics, ETH Zurich (Birkhauser-Verlag)
\bibitem{Y1}
Donati-Martin C, Shi Z and Yor M 2000
\newblock Ergod. Th. Dynam. Sys. {\bf 20} 709
\bibitem{MC}
Majumdar S N and Comtet A 2002
\newblock Phys. Rev. Lett. {\bf 89} 60601
\bibitem{JD}
Desbois J 2002
\newblock J. Phys. A: Math. Gen. {\bf 35} L673
\bibitem{bingham}
Bingham N H and Doney R A 1988
\newblock J. Appl. Prob. {\bf 25 } 120
\bibitem{RE}
Redner S 2001
\newblock{\it A Guide to First-Passage Processes} (Cambridge University Press)
\bibitem{A}
Arratia R 1979
\newblock{\it Coalescing Brownian motion on the line}, PHD thesis, Univ. of
Wisconsin (Madison, WI)
\bibitem{NOC}
O'Connell N and Unwin A 1992
\newblock Stochastic Processes and their Applications {\bf 43 } 291
\bibitem{B}
Biane P 1994
\newblock Stochastic Processes and their Applications {\bf 53 } 233
\bibitem{F}
Fisher M.E 1984
\newblock J. Stat. Phys. {\bf 34} 667
\bibitem{BA}
Ben-Avraham D 1988
\newblock J. Chem. Phys. {\bf 88} 941
\bibitem{M}
Monthus C 1996
\newblock Phys. Rev. E {\bf 54} 4844
\bibitem{DB}
De Blassie RD 1987
\newblock Prob. Th. Rel. Fields {\bf 75 }279
\bibitem{Y3}
Yor M Private communication
\bibitem{GRAD}
Gradstheyn I S and Ryzhik I M 1980
\newblock{\it Table of Integrals, Series, and Products} (New York: Academic)
%\bibitem{S}
%Spitzer F 1958
%\newblock Trans. Americ. Math. Soc. {\bf 87} 187
\end{thebibliography}
\end{document}