## Hitting distributions of geometric Brownian motion $\text{}$2000 MS Classification: Primary 60J65; Secondary 60J60. Key words and phrases: hyperbolic spaces, stable processes, Poisson kernel. Research partially supported by KBN grant 1 P03A 020 28 and RTN Harmonic Analysis and Related Problems contract HPRN-CT-2001-00273-HARP

### T. Byczkowski and M. Ryznar Institute of Mathematics, Wrocław University of Technology, Poland

Abstract
Let $\tau$  be the first hitting time of the point $1$  by the geometric Brownian motion $X\left(t\right)=xexp\left(B\left(t\right)-2\mu t\right)$  with drift $\mu ⩾0$  starting from $x>1$  . Here $B\left(t\right)$  is the Brownian motion starting from $0$  with $E〈0,B〈2,\left(t\right)=2t$  . We provide an integral formula for the density function of the stopped exponential functional $A\left(\tau \right)={\int }_{0}〈\tau ,X〈2,\left(t\right)dt$  and determine its asymptotic behaviour at infinity. Although we basically rely on methods developed in [BGS, the present paper also covers the case of arbitrary drifts $\mu ⩾0$  and provides a significant unification and extension of results of the above-mentioned paper. As a corollary we provide an integral formula and give asymptotic behaviour at infinity of the Poisson kernel for half-spaces for Brownian motion with drift in real hyperbolic spaces of arbitrary dimension.

1 Introduction

Let $B\left(t\right)$  be the standard Brownian motion starting from $0$  and normalized such that $E〈0,B\left(t\right)〈2,=2t$  . Consider the following linear SDE: $dX\left(t\right)=X\left(t\right)dB\left(t\right)-\left(2\mu -1\right)X\left(t\right)dt,X\left(0\right)=x>0,\mu ⩾0.$  The strong unique non-exploding solution is given by $X\left(t\right)=xexp\left(B\left(t\right)-2\mu t\right).$  The process $\left\{X\left(t\right);t⩾0\right\}$  is called a geometric Brownian motion or exponential Brownian motion and along with the additive functional from the process $X\left(t\right)$  , ${A}_{x}\left(t\right)={\int }_{0}〈t,X〈2,\left(t\right)dt={\int }_{0}〈t,x〈2,exp2\left(B\left(s\right)-2\mu s\right)ds$  is of primary interest in mathematical finance and insurance theory (see, e.g., [D, [GYor [Y2).
Also there is a connection of the above functional with the Brownian motion in hyperbolic halfspaces (see eg. [Y2, [AG, [BCF, [BCFYand Section 5).
The distribution of ${A}_{x}\left(t\right)$  for fixed $0\le t\le \infty$  has been a subject of study in substantial number of papers (see e.g. [B, [Y2, [AG, [M). In this paper we investigate the properties of the density function ${q}_{\mu }$  of the (stopped) additive functional ${A}_{x}\left(\tau \right)$  where $\tau$  is the first hitting time of the point $1$  by the process $X\left(t\right)$  (starting from $x>1$  ). From the strong Markov property it easily follows that the distribution of ${A}_{x}\left(\tau \right)$  is closely related to the distribution of ${A}_{x}\left(\infty \right)$  . The latter one is astonishingly simple; it is identical with the distribution of the random variable $x〈2,/4Z$  , where $Z$  is a $\Gamma \left(\mu ,1\right)$  -distributed random variable (with the density $\Gamma \left(\mu \right)〈-1,u〈\mu -1,e〈-u,$  ).
This fundamental result is due to D. Dufresne [D(see also [Y1, [Y2) and is of primary importance here. Knowing the Laplace transform of ${A}_{x}\left(\infty \right)$  one can easily derive (via the strong Markov property) the form of the Laplace transform of the random variable $A\left(\tau \right)$  as a suitable ratio of Bessel functions. For the reader's convenience we present this argument in Preliminaries. On the other hand the distributions of ${A}_{x}\left(\tau \right)$  and ${A}_{x}\left(\infty \right)$  are closely related to hitting times of Bessel processes. The main fact here is Lamperti's representation which reads that there exists a Bessel process $R〈\left(-\mu \right),$  with index $-\mu$  , starting from $x>1$  , such that the process $X\left(t\right)$  admits the following representation (see [Land Ex. 11.1.28 in [RY): $X\left(t\right)=R〈\left(-\mu \right),\left({A}_{x}\left(t\right)\right),t\ge 0.$  We refer the reader to [RYfor the account on Bessel processes (see also [GJY). From the Lamperti representation it follows immediately that ${A}_{x}\left(\tau \right)$  and ${A}_{x}\left(\infty \right)$  can be regarded as hitting times of $1$  and $0$  , respectively, of the Bessel process $R〈\left(-\mu \right),$  starting from $x>1$  . Also it is possible to relate ${A}_{x}\left(\tau \right)$  and ${A}_{x}\left(\infty \right)$  to last exit times for appropriate Bessel processes.
There is a vast existing literature on that subject. Such hitting or last exit times were studied in papers by Getoor [G, Getoor-Sharpe [GS, Kent [Kand Pitman-Yor [PY. In Getoor [G, Kent [Kand Getoor-Sharp [GSthe Laplace transform of the distribution of $A\left(\tau \right)$  is derived as a ratio of Bessel functions (see Preliminaries). For an exhausting discussion on that subject we refer to [GJY.
Our main focus in this paper is to provide an integral formula for the density of ${A}_{x}\left(\tau \right)$  (see Theorem  3.3 ). In the case $\mu =1/2$  this density is well known to be $1/2$  -stable subordinator.
Kent in [Kwrites that it is possible to obtain the explicit form of the density for $\mu =3/2$  but he does not provide any formula nor details on that.
At this point let us mention that various ratios of Bessel functions were proved to be completely monotonic functions hence they are Laplace transforms of probability distributions (see Ismail [I1,[I2, Ismail-Kelker [IK, Kent [K). For a survey on this theme see Pitman and Yor [PYand also [GY. Regarding the case which is considered in our paper Ismail and Kelker [IKshowed by purely analytical methods that it is an infinitely divisible distribution.
The main purpose of the paper is to obtain a suitable representation for the density function ${q}_{\mu }$  of the functional ${A}_{x}\left(\tau \right)$  along with its asymptotic properties, for arbitrary drift terms $\mu ⩾0$  .
Then we apply these results to derive an integral representation of the Poisson kernel for subspaces on real hyperbolic spaces, for hyperbolic Brownian motion with arbitrary drift, which extends and simplifies the results and proofs from [BGS.
The paper is organized as follows. In Preliminaries we collect basic information needed in the sequel.
In Section 3 we obtain a suitable representation of the density of the functional ${A}_{x}\left(\tau \right)$  . For this purpose we apply an integral representation given in [BGSfor ratio of Bessel functions.
In Section 4 we exhibit the exact asymptotics of the density of ${A}_{x}\left(\tau \right)$  at infinity for all drifts $\mu ⩾0$  . Again, we essentially follow the idea of [BGS. However, applying more direct probabilistic arguments, we are able to simplify our presentation considerably.
In Section 5 we show how to apply results obtained in preceding sections to obtain representation and asymptotic properties of Poisson kernel of subspaces on real hyperbolic spaces of arbitrary dimension, for Brownian motion with drift. Some of these asymptotic properties were studied in [BCFin the case of dimension 2, where it was shown that the distribution of Poisson kernel belongs to the stable domain of attraction. Our asymptotic results may be viewed as the extension of those obtained in [BCF. For a related result see also [BCFY.

2 Preliminaries

Let $0  and let ${\tau }_{a}$  be the first hitting time of the point $a$  by the geometric Brownian motion with drift $\mu ⩾0$  starting at $x$  : ${\tau }_{a}=inf\left\{t>0;xexp\left(B\left(t\right)-2\mu t\right)=a\right\}.$  The fact that $\tau <\infty$  a.e. follows from the property ${inf}_{t>0}B\left(t\right)=-\infty$  .
Further, define ${A}_{x}\left(t\right)={\int }_{0}〈t,exp2\left(B\left(t\right)-2\mu t\right)dt.$  By the strong Markov property of Brownian motion we obtain Basic relationship (for $\mu >0$  ).
Observe that $x〈2,exp2\left(W\left({\tau }_{a}\right)-2\mu {\tau }_{a}\right)=a〈2,$  , hence
 $\begin{array}{ccc}{A}_{x}\left(\infty \right)& =& x〈2,{\int }_{0}〈{\tau }_{a},exp2\left(W\left(s\right)-2\mu s\right)ds+x〈2,{\int }_{{\tau }_{a}}〈\infty ,exp2\left(W\left(s\right)-2\mu s\right)ds\end{array}$
 $\begin{array}{ccc}& =& {A}_{x}\left({\tau }_{a}\right)+x〈2,exp2\left(W\left({\tau }_{a}\right)-2\mu {\tau }_{a}\right){\int }_{0}〈\infty ,exp2\left(W\left(s+{\tau }_{a}\right)-W\left({\tau }_{a}\right)-2\mu s\right)ds\end{array}$
 $\begin{array}{ccc}& =& {A}_{x}\left({\tau }_{a}\right)+{A}_{a}^{\prime }\left(\infty \right),\end{array}$ (1)
where ${A}_{a}^{\prime }\left(\infty \right)$  is a copy of ${A}_{a}\left(\infty \right)$  , independent from ${A}_{x}\left({\tau }_{a}\right)$  .
Dufresne [D(see also Getoor [G, Kent [K, Getoor-Sharpe [GS, where the result is given in the context of Bessel processes) showed that the density and the Laplace transform of ${A}_{1}\left(\infty \right)$  have the following form:
 $\begin{array}{c}{h}_{\mu }\left(t\right)=2〈-2\mu ,\frac{}{}\Gamma \left(\mu \right)e〈-1/4t,\frac{}{}t〈1+\mu ,,\end{array}$ (2)
 $\begin{array}{c}\mathcal{ℒ}\left\{{A}_{1}\left(\infty \right)\right\}\left(r〈2,\right)=\mathcal{ℒ}\left\{1/4Z\right\}\left(r〈2,\right)=2\left(r/2\right)〈\mu ,\frac{}{}\Gamma \left(\mu \right){K}_{\mu }\left(r\right),\mu >0.\end{array}$ (3)
From this and elementary properties of Laplace transform one immediately obtains
 $\begin{array}{c}\mathcal{ℒ}\left\{{A}_{x}\left({\tau }_{a}\right)\right\}\left(r〈2,\right)=\left(x\frac{}{}a\right)〈\mu ,{K}_{\mu }\left(xr\right)\frac{}{}{K}_{\mu }\left(ar\right),\end{array}$ (4)
where ${K}_{\mu }$  is the modified Bessel function of second type with index $\mu$  . The above Laplace transform appears in many papers (see eg. Kent [K, Getoor-Sharpe [GS, Ismail-Kelker [IK) and here we followed a derivation from [BCF. As mentioned in the Introduction we present it for the readers convenience.
From the continuity of both sides of ( 4 ) with respect to $\mu ⩾0$  we obtain that the above formula remains valid also for $\mu =0$  .
Let us remark that for $\mu =0$  the right-hand side of ( 4 ) gives also the classical formula ([S) for the Laplace transform of Brownian motion hitting time ${T}_{a}$  of the centered circle with radius $a$  in $\mathbb{R}〈2,$  from the point $y\in \mathbb{R}〈2,$  such that $|y|=x>a$  .
By the formula ( 4 ) follows directly that for $t>0$  we have ${A}_{ta,tx}\left({\tau }_{ta}\right)\stackrel{d}{=}t〈2,{A}_{a,x}\left({\tau }_{a}\right),$  where the symbol $\stackrel{d}{=}$  denotes the equality of distribution.
Therefore, from now on we may and do assume that $a=1$  and $x>1$  is fixed. We write $\tau$  instead of ${\tau }_{1}$  and $A\left(\tau \right)$  instead of ${A}_{1,x}\left({\tau }_{1}\right)$  .
We conclude this section with the following technical lemma
Lemma 2.1.
 $\begin{array}{c}{\int }_{0}〈\infty ,e〈-y〈2,/4t,{h}_{s}\left(t/\lambda 〈2,\right)dt=\lambda 〈2s+2,\frac{}{}\left(\lambda 〈2,+y〈2,\right)〈s,.\end{array}$ (5)
• Proof. Indeed, we obtain  $\begin{array}{ccc}& & {\int }_{0}〈\infty ,e〈-y〈2,/4t,e〈-\lambda 〈2,/4t,\lambda 〈2s+2,dt\frac{}{}t〈1+s,=\lambda 〈2s+2,\frac{}{}\left(\lambda 〈2,+y〈2,\right)〈s,{\int }_{0}〈\infty ,e〈-\left(y〈2,+\lambda 〈2,\right)/4t,dt/\left(y〈2,+\lambda 〈2,\right)\frac{}{}\left(t/\left(y〈2,+\lambda 〈2,\right)\right)〈1+s,\end{array}$
 $\begin{array}{ccc}& =& 2〈2s,\lambda 〈2s+2,\frac{}{}\left(\lambda 〈2,+y〈2,\right)〈s,{\int }_{0}〈\infty ,\Gamma \left(s\right){h}_{s}\left(u\right)du=2〈2s,\Gamma \left(s\right)\lambda 〈2s+2,\frac{}{}\left(\lambda 〈2,+y〈2,\right)〈s,.\end{array}$
Throughout the paper we use the following convention: by $c,C$  we always denote nonnegative constants which may depend on other constant parameters only. The value of $c$  or $C$  may change from line to line in a chain of estimates.
The notion $p\left(u\right)\approx q\left(u\right),u\to {u}_{0}$  means that the ratio $q/p\to 1$  when $u\to {u}_{0}$  .

3 Representation of density of $A\left(\tau \right)$

In this section we give a representation formula for the density function of functional $A\left(\tau \right)$  , for arbitrary $\mu ⩾0$  .
From now on we use the following notation, partially introduced in the preceding Section: $x>1,a=1,\lambda =x-1.$  and we denote by ${q}_{\mu }$  the density function of the functional $A\left(\tau \right)$  .
We begin with stating a more general version of a lemma, taken from [BGS. The proof is identical and is omitted.
Lemma 3.1. Let $\mu ⩾0$  . Suppose that $Q\left(z\right)=z-\left(\mu 〈2,-1/4\right)\lambda \frac{}{}2x,z\in \mathbb{C}.$  Define ${F}_{\lambda }\left(z\right)$  by the following formula:
 $\begin{array}{c}\lambda {F}_{\lambda }\left(z\right)=ze〈\lambda z,x〈\mu ,{K}_{\mu }\left(xz\right)-x〈\mu -1\frac{}{}2,Q\left(z\right){K}_{\mu }\left(z\right)\frac{}{}{K}_{\mu }\left(z\right).\end{array}$ (6)
Then
 $\begin{array}{c}{F}_{\lambda }\left(z\right)=O\left(z〈-1,\right),z\to \infty .\end{array}$ (7)
and there exists a function ${w}_{\lambda }$  such that
 $\begin{array}{c}{F}_{\lambda }\left(z\right)={\int }_{0}〈\infty ,e〈-zv,{w}_{\lambda }\left(v\right)dv.\end{array}$ (8)
Moreover,
 $\begin{array}{c}x〈\mu -1/2,\left(\mu 〈2,-1/4\right)/2x={\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)dv,\end{array}$ (9)
and, for $\mu >1/2$  ,
 $\begin{array}{c}2x〈\mu -1/2,={\int }_{0}〈\infty ,\kappa {w}_{\lambda }\left(v\right)dv,\end{array}$ (10)
where $\kappa =\left(\lambda +v\right)〈2,-\lambda 〈2,=v\left(2\lambda +v\right).$  For $\mu =1/2$  we have ${F}_{\lambda }\left(z\right)\equiv 0$  .
An explicit formula for the function ${w}_{\lambda }$  will be provided in the sequel.
The following formula is crucial for our considerations:
Lemma 3.2. Let $\phi :\left[0,\infty \right)\to \mathbb{R}〈+,$  belong to $L〈1,\left(0,\infty \right)$  such that its Laplace transform $\stackrel{^}{\phi }$  has the following property $t〈-1/2,\stackrel{^}{\phi }\left(t\right)\in L〈1,\left(0,\infty \right).$  Then we have
 $\begin{array}{c}{\int }_{0}〈\infty ,\stackrel{^}{\phi }\left(r〈2,\right)cos\left(ry\right)dr=\sqrt{\pi }/2{\int }_{0}〈\infty ,e〈-y〈2,/4t,\phi \left(t\right)dt\frac{}{}\sqrt{t}.\end{array}$ (11)
• Proof. Let $A$  be a random variable with absolutely continuous distribution with the density function $\phi$  and let $W\left(t\right)$  be Brownian Motion starting from $0$  (such that $EW〈2,\left(t\right)=2t$  ), independent from $A$  . It is easy to see that the value of the Laplace transform of $A$  at the point $r〈2,$  , that is $\stackrel{^}{\phi }\left(r〈2,\right)$  is equal to the Fourier (or cosine transform) of $W\left(A\right)$  at the point $r$  .
Observe that our assumption assures that this Fourier transform belongs to $L〈1,\left(0,\infty \right)$  . Thus, the left-hand side of ( 11 ) is the inversion formula applied for the Fourier transform and gives the density of $W\left(A\right)$  . The right-hand side results from the direct computation of this density, taking into account the independence of $W$  and $A$  and the particular form of the (gaussian) density of $W\left(t\right)$  .
We are ready to state our representation formula.
Theorem 3.3. For $\mu ⩾0$  we have
 $\begin{array}{c}{q}_{\mu }\left(t\right)=\lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi t}\left(x〈\mu -1/2,/2t+{\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-1\right){w}_{\lambda }\left(v\right)dv\right).\end{array}$ (12)
For $\mu >1/2$  we have
 $\begin{array}{c}{q}_{\mu }\left(t\right)=\lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi t}{\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-1+\kappa /4t\right){w}_{\lambda }\left(v\right)dv.\end{array}$ (13)
• Proof. The proof relies on application of the formula ( 11 ) for the function $\stackrel{^}{\phi }\left(r〈2,\right)=x〈\mu ,{K}_{\mu }\left(xr\right)\frac{}{}{K}_{\mu }\left(r\right).$  Note that by ( 6 ) and ( 9 ) it follows that  $\begin{array}{ccc}rx〈\mu ,{K}_{\mu }\left(xr\right)\frac{}{}{K}_{\mu }\left(r\right)& =& e〈-\lambda r,\lambda {F}_{\lambda }\left(r\right)+e〈-\lambda r,x〈\mu -1/2,Q\left(r\right)\end{array}$
 $\begin{array}{ccc}& =& e〈-\lambda r,x〈\mu -1/2,r+e〈-\lambda r,\lambda {F}_{\lambda }\left(r\right)-e〈-\lambda r,x〈\mu -1/2,\left(\mu 〈2,-1/4\right)\lambda /2x\end{array}$
 $\begin{array}{ccc}& =& e〈-\lambda r,x〈\mu -1/2,r+\lambda e〈-\lambda r,{\int }_{0}〈\infty ,\left(e〈-rv,-1\right){w}_{\lambda }\left(v\right)dv.\end{array}$
We recall that $Q\left(r\right)=r-\left(\mu 〈2,-1/4\right)\lambda /2x$  and ${F}_{\lambda }\left(r\right)={\int }_{0}〈\infty ,e〈-rv,{w}_{\lambda }\left(v\right)dv$  . To simplify the proof we introduce new notation. Namely, denote ${w}_{\lambda }〈#,\left(v\right)=-{\int }_{v}〈\infty ,{w}_{\lambda }\left(v\right)dv,\text{that is}d{w}_{\lambda }〈#,\left(v\right)\frac{}{}dv={w}_{\lambda }\left(v\right).$  Then, by integration by parts we obtain  $\begin{array}{ccc}{\int }_{0}〈\infty ,\left(e〈-rv,-1\right){w}_{\lambda }\left(v\right)dv& =& \left(e〈-rv,-1\right){w}_{\lambda }〈#,\left(v\right){|}_{0}〈\infty ,+r{\int }_{0}〈\infty ,e〈-rv,{w}_{\lambda }〈#,\left(v\right)dv\end{array}$
 $\begin{array}{ccc}& =& r{\int }_{0}〈\infty ,e〈-rv,{w}_{\lambda }〈#,\left(v\right)dv.\end{array}$
Thus, the left-hand side of the formula ( 11 ) is now of the form  $\begin{array}{ccc}& & {\int }_{0}〈\infty ,rx〈\mu ,{K}_{\mu }\left(xr\right)\frac{}{}{K}_{\mu }\left(r\right)cos\left(ry\right)/rdr\end{array}$
 $\begin{array}{ccc}& =& x〈\mu -1/2,{\int }_{0}〈\infty ,e〈-\lambda r,cos\left(ry\right)dr\end{array}$
 $\begin{array}{ccc}& +& {\int }_{0}〈\infty ,\lambda e〈-\lambda r,\left\{{\int }_{0}〈\infty ,\left(e〈-rv,-1\right){w}_{\lambda }\left(v\right)dv\right\}cos\left(ry\right)/rdr\end{array}$
 $\begin{array}{ccc}& =& x〈\mu -1/2,{\int }_{0}〈\infty ,e〈-\lambda r,cos\left(ry\right)dr\end{array}$
 $\begin{array}{ccc}& +& {\int }_{0}〈\infty ,\lambda e〈-\lambda r,\left\{{\int }_{0}〈\infty ,e〈-rv,{w}_{\lambda }〈#,\left(v\right)dv\right\}cos\left(ry\right)dr\end{array}$
 $\begin{array}{ccc}& =& {H}_{\mu }〈\left(1\right),\left(y\right)+{H}_{\mu }〈\left(2\right),\left(y\right)={H}_{\mu }\left(y\right).\end{array}$
Using the standard formula for the Laplace transform of the cosine function we obtain ${H}_{\mu }〈\left(1\right),\left(y\right)=x〈\mu -1/2,\lambda \frac{}{}\lambda 〈2,+y〈2,,$  and ${H}_{\mu }〈\left(2\right),\left(y\right)=\lambda {\int }_{0}〈\infty ,\left(\lambda +v\right){w}_{\lambda }〈#,\left(v\right)\frac{}{}\left(\lambda +v\right)〈2,+y〈2,dv.$  By the form of ${H}_{\mu }〈\left(1\right),\left(y\right)$  , ( 5 ) (applied for $s=1$  ), ( 9 ) and ( 10 ) we obtain ${H}_{\mu }〈\left(1\right),\left(y\right)=x〈\mu -1/2,\lambda \frac{}{}2〈2,{\int }_{0}〈\infty ,e〈-y〈2,/4t,e〈-\lambda 〈2,/4t,dt\frac{}{}t〈2,.$  Analogously, for ${H}_{\mu }〈\left(2\right),$  we obtain  $\begin{array}{ccc}{H}_{\mu }〈\left(2\right),\left(y\right)& =& \lambda \frac{}{}2〈2,{\int }_{0}〈\infty ,e〈-y〈2,/4t,\left({\int }_{0}〈\infty ,\left(\lambda +v\right){w}_{\lambda }〈#,\left(v\right)e〈-\left(\lambda +v\right)〈2,/4t,dv\right)dt\frac{}{}t〈2,\end{array}$
 $\begin{array}{ccc}& =& \lambda \frac{}{}2{\int }_{0}〈\infty ,e〈-y〈2,/4t,\left({\int }_{0}〈\infty ,d\left(-e〈-\left(\lambda +v\right)〈2,/4t,\right)\frac{}{}dv{w}_{\lambda }〈#,\left(v\right)dv\right)dt\frac{}{}t.\end{array}$
 $\begin{array}{ccc}& & \end{array}$
Observe now that  $\begin{array}{ccc}{\int }_{0}〈\infty ,d\left(-e〈-\left(\lambda +v\right)〈2,/4t,\right)\frac{}{}dv{w}_{\lambda }〈#,\left(v\right)dv& =& -e〈-\left(\lambda +v\right)〈2,/4t,{w}_{\lambda }〈#,\left(v\right){|}_{0}〈\infty ,+{\int }_{0}〈\infty ,e〈-\left(\lambda +v\right)〈2,/4t,d{w}_{\lambda }〈#,\left(v\right)\frac{}{}dvdv\end{array}$
 $\begin{array}{ccc}& =& e〈-\lambda 〈2,/4t,{w}_{\lambda }〈#,\left(0\right)+{\int }_{0}〈\infty ,e〈-\left(\lambda +v\right)〈2,/4t,{w}_{\lambda }\left(v\right)dv.\end{array}$
We also have ${w}_{\lambda }〈#,\left(0\right)=-{\int }_{0}〈\infty ,d{w}_{\lambda }〈#,\left(v\right)\frac{}{}dvdv=-{\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)dv.$  By the above identities and the form of ${H}_{\mu }〈\left(2\right),$  we obtain ${H}_{\mu }〈\left(2\right),\left(y\right)=\lambda \frac{}{}2{\int }_{0}〈\infty ,e〈-y〈2,/4t,\left({\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-1\right){w}_{\lambda }\left(v\right)dv\right)dt\frac{}{}t.$  Combining the above identities we obtain that the left-hand side of ( 11 ) takes the form ${H}_{\mu }\left(y\right)=\lambda \frac{}{}2{\int }_{0}〈\infty ,e〈-y〈2,/4t,e〈-\lambda 〈2,/4t,\left(x〈\mu -1/2,/2t+{\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-1\right){w}_{\lambda }\left(v\right)dv\right)dt\frac{}{}t.$  Taking into account the right-hand side of ( 11 ) and continuity of ${q}_{\mu }$  with respect to $t$  (see properties of ${w}_{\lambda }$  below), we obtain ( 12 ). When $\mu >1/2$  , then using ( 10 ) we obtain ( 13 ).
Below we give a description of the function ${w}_{\lambda }$  . We rely here on results contained in [BGS.
The formulas depend on the zeros of the function ${K}_{\mu }\left(z\right)$  .
Even if in general the values of these zeros are not given explicitly, we are able to prove some important properties (as boundedeness or asymptotics) of ${w}_{\lambda }$  , which are essential in applications. Moreover, for some values of $\mu$  we provide explicit formulas as well (see Corollary  3.5 ).
The function ${K}_{\mu }\left(z\right)$  extends to an entire function when $\mu -1/2$  is an integer and has a holomorphic extension to $\mathbb{C}\\left(-\infty ,0\right]$  when $\mu -1/2$  is not an integer. Denote the set of zeros of the function ${K}_{\mu }\left(z\right)$  by $Z=\left\{{z}_{1},...,{z}_{{k}_{\mu }}\right\}$  . We give some information about these zeros (cf. [E, p. 62) needed in the sequel. Recall that ${k}_{\mu }=\mu -1/2$  when $\mu -1/2\in \mathbb{N}$  . For $\mu -1/2/\in \mathbb{N}$  , ${k}_{\mu }$  is the even number closest to $\mu -1/2$  . In particular, for $0⩽\mu <3/2$  we have ${k}_{\mu }=0$  ; for $\mu =2$  and $3$  we have ${k}_{\mu }=2$  . The functions ${K}_{\mu }$  and ${K}_{\mu -1}$  have no common zeros.
As in [BGS, we need an additional notation to describe the function ${w}_{\lambda }$  . Define for $\mu >0$
 $\begin{array}{c}{w}_{1,\lambda }\left(v\right)=-x〈\mu ,\frac{}{}\lambda {\sum }_{i=1}〈{k}_{\mu },{z}_{i}e〈\lambda {z}_{i},{K}_{\mu }\left(x{z}_{i}\right)\frac{}{}{K}_{\mu -1}\left({z}_{i}\right)e〈{z}_{i}v,.\end{array}$ (14)
When $\mu +1/2/\in \mathbb{N}$  and $\mu \ge 0$  we define
 $\begin{array}{c}{w}_{2,\lambda }\left(v\right)=-cos\left(\pi \mu \right)x〈\mu ,\frac{}{}\lambda {\int }_{0}〈\infty ,{I}_{\mu }\left(xu\right){K}_{\mu }\left(u\right)-{I}_{\mu }\left(u\right){K}_{\mu }\left(xu\right)\frac{}{}cos〈2,\left(\pi \mu \right){K}_{\mu }〈2,\left(u\right)+\left(\pi {I}_{\mu }\left(u\right)+sin\left(\pi \mu \right){K}_{\mu }\left(u\right)\right)〈2,e〈-\lambda u,e〈-vu,udu.\end{array}$ (15)
We now formulate our representation theorem for the function ${w}_{\lambda }$  . The proof of the main part is the same as in [BGSand is omitted; we only show asymptotic properties of the function ${w}_{2,\lambda }$  . For $\mu =0$  this is new; behaviour for $\mu =\left(n-1\right)/2$  was shown in [BGS. Nevertheless, we present here a new and unified proof based on tauberian theorems.
Theorem 3.4. In the case $\mu -1/2\in \mathbb{N}$  ${w}_{\lambda }\left(v\right)={w}_{1,\lambda }\left(v\right);$  while, in the case when $\mu -1/2/\in \mathbb{N}$  ${w}_{\lambda }\left(v\right)={w}_{1,\lambda }\left(v\right)+{w}_{2,\lambda }\left(v\right).$  Moreover, we have ${sup}_{v⩾0}|{w}_{\lambda }\left(v\right)|<\infty ,$  and $-cos\left(\pi \mu \right){w}_{2,\lambda }\left(v\right)⩾0,v⩾0,\left(\mu -1/2/\in \mathbb{N}\right);$  ${\int }_{0}〈\infty ,v〈k,|{w}_{1,\lambda }\left(v\right)|dv<\infty ,k=1,2,\dots ;$  ${lim}_{v\to \infty }v〈k,{w}_{1,\lambda }\left(v\right)=0,k=1,2,\dots ;$  ${lim}_{v\to \infty }v〈2\mu +2,{w}_{2,\lambda }\left(v\right)=-cos\left(\pi \mu \right)\Gamma \left(2\mu +2\right)\frac{}{}2〈2\mu -3,\Gamma \left(\mu \right)\Gamma \left(\mu +1\right)x〈2\mu ,-1\frac{}{}\lambda ,\left(\mu -1/2/\in \mathbb{N},\mu >0\right);$  ${lim}_{v\to \infty }\left(vlogv\right)〈2,{w}_{2,\lambda }\left(v\right)=-1\frac{}{}2\lambda logx,\text{for}\mu =0\text{}.$
• Proof. Denote  $\begin{array}{c}{h}_{\mu ,\lambda }\left(u\right)={I}_{\mu }\left(xu\right){K}_{\mu }\left(u\right)-{I}_{\mu }\left(u\right){K}_{\mu }\left(xu\right)\frac{}{}cos〈2,\left(\pi \mu \right){K}_{\mu }〈2,\left(u\right)+\left(\pi {I}_{\mu }\left(u\right)+sin\left(\pi \mu \right){K}_{\mu }\left(u\right)\right)〈2,e〈-\lambda u,u.\end{array}$ (16)
Observe that the function ${h}_{\mu ,\lambda }\left(u\right)$  is non-negative ( ${I}_{\mu }\left(u\right)/{K}_{\mu }\left(u\right)$  is increasing for $u>0$  ) and ${w}_{2,\lambda }\left(v\right)$  is the Laplace transform of $-cos\left(\pi \mu \right)x〈\mu ,\frac{}{}\lambda {h}_{\mu ,\lambda }$  at $v$  . We claim that for $u\to 0+$  we have the following asymptotics for ${h}_{\mu ,\lambda }$  :  $\begin{array}{c}{h}_{\mu ,\lambda }\left(u\right)\approx \left\{\begin{array}{cc}x〈\mu ,\frac{{c}_{\mu }}{{c}_{\mu }^{\prime }}\left(1-x〈-2\mu ,\right)u〈2\mu +1,,& \text{for}\mu >0\text{},\\ ulogx\left(logu\right)〈-2,,& \text{for}\mu =0\text{}.\end{array}\end{array}$ (17)
Applying Karamata's Tauberian theorem (see, e.g. [Fe) we obtain that the asymptotic behaviour of ${w}_{2,\lambda }\left(v\right)$  is ${lim}_{v\to \infty }{w}_{2,\lambda }\left(v\right)\left({\int }_{0}〈1/v,{h}_{\mu ,\lambda }\left(u\right)\right)〈-1,=-cos\left(\pi \mu \right)\frac{}{}\lambda \Gamma \left(2+2\mu \right)x〈\mu ,.$  This together with ( 17 ) implies:
 $\begin{array}{ccc}& & {lim}_{v\to \infty }v〈2\mu +2,{w}_{2,\lambda }\left(v\right)=-cos\left(\pi \mu \right)\Gamma \left(2\mu +2\right)\frac{}{}2〈2\mu -3,\Gamma \left(\mu \right)\Gamma \left(\mu +1\right)x〈2\mu ,-1\frac{}{}\lambda ,\left(\mu -1/2/\in \mathbb{N},\mu >0\right),\end{array}$
 $\begin{array}{ccc}& & {lim}_{v\to \infty }\left(vlogv\right)〈2,{w}_{2,\lambda }\left(v\right)=-1\frac{}{}2\lambda logx,\mu =0.\end{array}$
To prove ( 17 ) we apply the following asymptotics. When $u\to 0$  we have:
 $\begin{array}{ccc}{I}_{\mu }\left(u\right)& \approx & {c}_{\mu }u〈\mu ,,{K}_{\mu }\left(u\right)\approx {c}_{\mu }^{\prime }u〈-\mu ,;\end{array}$ (18)
 $\begin{array}{ccc}{I}_{0}\left(u\right)& =& 1+o\left(1\right),{K}_{0}\left(u\right)=log\left(2/u\right){I}_{0}\left(u\right)+\psi \left(1\right)+o\left(1\right),\end{array}$ (19)
with ${c}_{\mu }=2〈-\mu ,/\Gamma \left(\mu +1\right)$  , ${c}_{\mu }^{\prime }=2〈\mu -1,\Gamma \left(\mu \right)$  and where $\psi$  is the Euler function. Then  $\begin{array}{ccc}e〈\lambda u,u〈-1,{h}_{\mu ,\lambda }\left(u\right)& =& {I}_{\mu }\left(xu\right){K}_{\mu }\left(u\right)\frac{}{}{K}_{\mu }〈2,\left(u\right)+\left(\pi {I}_{\mu }\left(u\right)\right)〈2,+2\pi sin\left(\pi \mu \right){K}_{\mu }\left(u\right){I}_{\mu }\left(u\right)\left(1-{I}_{\mu }\left(u\right){K}_{\mu }\left(xu\right)\frac{}{}{I}_{\mu }\left(xu\right){K}_{\mu }\left(u\right)\right)\end{array}$
 $\begin{array}{ccc}& \approx & {I}_{\mu }\left(xu\right)\frac{}{}{K}_{\mu }\left(u\right)\left(1-{I}_{\mu }\left(u\right){K}_{\mu }\left(xu\right)\frac{}{}{I}_{\mu }\left(xu\right){K}_{\mu }\left(u\right)\right)\end{array}$
 $\begin{array}{ccc}& \approx & \left\{\begin{array}{cc}u〈2\mu ,x〈\mu ,\frac{{c}_{\mu }}{{c}_{\mu }^{\prime }}\left(1-x〈-2\mu ,\right),& \text{for}\mu >0\text{},\\ logx\frac{}{}\left(logu\right)〈2,,& \text{for}\mu =0\text{}.\end{array}\end{array}$
Examples. To illustrate representation theory developed so far we write down explicit integral formulas for the density ${q}_{\mu }$  in some special cases of $\mu$  . All formulas appearing here follow directly from Theorems  3.3 and  3.4 . If $0⩽\mu <3/2$  then ${w}_{\lambda }={w}_{2,\lambda }$  and functions ${w}_{\lambda }$  have constant sign. For $\mu =0$  and $\mu =1$  the function ${w}_{\lambda }$  has simpler form, which we exhibit here. If $\mu =1/2$  then the function ${F}_{\lambda }=0$  and ${q}_{\mu }\left(t\right)$  reduces to the standard $1/2$  -stable subordinator. For $\mu +1/2\in \mathbb{N}$  we have, in turn that ${w}_{\lambda }={w}_{1,\lambda }$  and the form of ${w}_{1,\lambda }$  can be computed calculating residues of simple rational functions (see the formula for ${w}_{1,\lambda }$  or calculations in [BGS). Again, we write the explicit form of ${w}_{\lambda }$  for $\mu =3/2$  and $\mu =5/2$  .
Corollary 3.5. If $\mu =0$  then $-{w}_{\lambda }\left(v\right)=1\frac{}{}\lambda {\int }_{0}〈\infty ,{I}_{0}\left(xu\right){K}_{0}\left(u\right)-{K}_{0}\left(xu\right){I}_{0}\left(u\right)\frac{}{}{K}_{0}〈2,\left(u\right)+\pi 〈2,{I}_{0}〈2,\left(u\right)e〈-u\lambda ,e〈-vu,udu,$  and ${q}_{\mu }\left(t\right)=\lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi t}\left(\left(\lambda +1\right)〈-1/2,/2t+{\int }_{0}〈\infty ,\left(1-e〈-\kappa /4t,\right)\left(-{w}_{\lambda }\left(v\right)\right)dv\right).$  If $\mu =1/2$  then ${q}_{\mu }\left(t\right)=\lambda e〈-\lambda 〈2,/4t,\frac{}{}2\sqrt{\pi t〈3,}.$  If $\mu =1$  then ${w}_{\lambda }\left(v\right)=\lambda +1\frac{}{}\lambda {\int }_{0}〈\infty ,{I}_{1}\left(xu\right){K}_{1}\left(u\right)-{K}_{1}\left(xu\right){I}_{1}\left(u\right)\frac{}{}{K}_{1}〈2,\left(u\right)+\pi 〈2,{I}_{1}〈2,\left(u\right)e〈-u\lambda ,e〈-vu,udu,$  and ${q}_{\mu }\left(t\right)=\lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi t}{\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-1+\kappa /4t\right){w}_{\lambda }\left(v\right)dv.$  If $\mu =3/2$  then ${w}_{\lambda }\left(v\right)=e〈-v,$  and ${q}_{\mu }\left(t\right)=\lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi t}{\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-1+\kappa /4t\right)e〈-v,dv.$  If $\mu =5/2$  then ${w}_{\lambda }\left(v\right)=3e〈-3v/2,\left[\left(2\lambda +1\right)cos\left(\sqrt{3}v/2\right)+\sqrt{3}sin\left(\sqrt{3}v/2\right)\right],$  and ${q}_{\mu }\left(t\right)=\lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi t}{\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-1+\kappa /4t\right){w}_{\lambda }\left(v\right)dv.$

4 Asymptotic behaviour of $A\left(\tau \right)$

In this section we prove the following
Theorem 4.1. The density ${q}_{\mu }$  of $A\left(\tau \right)$  satisfies:
${lim}_{t\to \infty }t〈\mu +1,{q}_{\mu }\left(t\right)={C}_{\mu },\text{if}\mu >0\text{},$  ${lim}_{t\to \infty }\left(logt\right)〈2,t{q}_{\mu }\left(t\right)={C}_{0},\text{if}\mu =0\text{},$  for some positive ${C}_{\mu }$  .
Lemma 4.2. Let $\mu >0$  . There exists a constant $C>1$  such that $C〈-1,t〈-\mu ,\le P\left(A\left(\tau \right)>t\right)\le Ct〈-\mu ,,t>1.$
• Proof. From ( 1 ) we infer that $\left(x>1,a=1\right)$  :
$P\left(A\left(\infty \right)>t\right)\le P\left(A\left(\tau \right)>\left(1-x〈-1,\right)tx〈2,\right)+P\left(A\left(\infty \right)>xt\right),$  which implies  $\begin{array}{c}P\left(t\le A\left(\infty \right)\le xt\right)\le P\left(A\left(\tau \right)>\left(x-1\right)xt\right).\end{array}$ (20)
Moreover,  $\begin{array}{c}P\left(A\left(\tau \right)>x〈2,t\right)\le P\left(A\left(\infty \right)>t\right).\end{array}$ (21)
Now the lemma follows from ( 20 ) and ( 21 ) since $P\left(A\left(\infty \right)>t\right)\approx ct〈-\mu ,$  for some $c>0$  by ( 2 ).
Recall that $\kappa =\left(\lambda +v\right)〈2,-\lambda 〈2,$  .
Lemma 4.3. Let $m\in \mathbb{N}$  be such that $2\le m\le \mu +1/2$  . Then
 $\begin{array}{c}{lim}_{t\to \infty }t〈m,{\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)\left(e〈-\kappa /4t,-{\sum }_{0⩽j⩽m-1}\left(-1\right)〈j,1\frac{}{}j!\left(\frac{\kappa }{4t}\right)〈j,\right)dv=\frac{\left(-1\right)〈m,}{4〈m,m!}{\int }_{0}〈\infty ,\kappa 〈m,{w}_{\lambda }\left(v\right)dv={C}_{m}.\end{array}$ (22)
Moreover, ${C}_{m}=0$  , for $2\le m<\mu +1/2$  , and, in the case $m=\mu +1/2\in \mathbb{N}$  , we have ${C}_{m}>0$  .
• Proof. Denote $\psi \left(t,m\right)=t〈m,{w}_{\lambda }\left(v\right)\left(e〈-\kappa /4t,-{\sum }_{0⩽j⩽m-1}\left(-1\right)〈j,1\frac{}{}j!\left(\frac{\kappa }{4t}\right)〈j,\right).$  By elementary calculations $|e〈-\kappa /4t,-{\sum }_{0⩽j⩽m-1}\left(-1\right)〈j,1\frac{}{}j!\left(\frac{\kappa }{4t}\right)〈j,|\le 1\frac{}{}m!\left(\frac{\kappa }{4t}\right)〈m,.$  Hence $|\psi \left(t,m\right)|\le |{w}_{\lambda }\left(v\right)|\kappa 〈m,\frac{}{}m!.$  Under the assumption on $m$  the function $|{w}_{\lambda }\left(v\right)|\kappa 〈m,$  is integrable so the formula ( 22 ) follows from the bounded convergence theorem.
Suppose that there exists ${C}_{m}\ne 0$  , with $m$  having properties as above, and denote ${m}_{0}=inf\left\{m\in \mathbb{N}:2\le m⩽\mu +1/2,{C}_{m}\ne 0\right\}.$  Then from the first part of the proof we have ${lim}_{t\to \infty }t〈{m}_{0}+1/2,{q}_{\mu }\left(t\right)={C}_{{m}_{0}}>0,$  which implies that ${lim}_{t\to \infty }t〈{m}_{0}-1/2,P\left(A\left(\tau \right)>t\right)={C}_{{m}_{0}}\left({m}_{0}-1/2\right)〈-1,.$  From Lemma  4.2 we infer that ${C}_{{m}_{0}}>0$  if and only if ${m}_{0}-1/2=\mu$  . In particular, we then have $\mu +1/2\in \mathbb{N}$  . We also obtained that if $m<\mu +1/2$  then ${C}_{m}=0$  . This completes the proof of the lemma.
Remark. The above lemma yields, in particular, that for $m\in \mathbb{N},2⩽m<\mu +1/2$  ${\int }_{0}〈\infty ,\kappa 〈m,{w}_{\lambda }\left(v\right)dv=0.$  Thus, the representation formula for the density ${q}_{\mu }$  can be written for $\mu ⩾1/2$  as follows:
 $\begin{array}{c}{q}_{\mu }\left(t\right)=\lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi t}{\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-{\sum }_{0⩽j⩽l}\left(-1\right)〈j,1\frac{}{}j!\left(\frac{\kappa }{4t}\right)〈j,\right){w}_{\lambda }\left(v\right)dv,\end{array}$ (23)
where $l=\left[\mu +1/2\right]$  , if $\mu -1/2/\in \mathbb{N}$  , and $l=\mu -1/2$  otherwise.
We now prove our theorem.
Proof of Theorem  4.1
• Proof. For $\mu =1/2$  the density ${q}_{\mu }$  has a particularly simple form (see Corollary  3.5 ) and the theorem clearly holds true. Hence, we assume throughout the remainder of the proof that $\mu \ne 1/2$  . Next, if $\mu -1/2\in \mathbb{N}$  then the Remark above together with Lemma  4.3 yield our theorem at once.
Thus, we assume for what follows that $\mu -1/2/\in \mathbb{N}$  and let $l=\left[\mu +1/2\right]$  . Denote $I\left(t\right)={\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)t〈\mu +1/2,\left(e〈-\kappa /4t,-{\sum }_{0⩽j⩽l}\left(-1\right)〈j,1\frac{}{}j!\left(\frac{\kappa }{4t}\right)〈j,\right)dv.$  We prove that  $\begin{array}{c}{lim}_{t\to \infty }I\left(t\right)=C>0,\text{if}\mu >0;\end{array}$ (24)
and  $\begin{array}{c}{lim}_{t\to \infty }\left(logt\right)〈2,I\left(t\right)=C>0,\text{if}\mu =0.\end{array}$ (25)
Applying change of variable $\kappa =4st$  we obtain $v=\sqrt{4st+\lambda 〈2,}-\lambda ,dv=\frac{2t}{\sqrt{4st+\lambda 〈2,}}ds,$  so $I\left(t\right)={\int }_{0}〈\infty ,{\psi }_{\lambda }\left(s,t\right)\left(e〈-s,-{\sum }_{0⩽j⩽l}\left(-1\right)〈j,s〈j,\frac{}{}j!\right)ds,$  where ${\psi }_{\lambda }\left(s,t\right)={w}_{\lambda }\left(\sqrt{4st+\lambda 〈2,}-\lambda \right)\frac{2t〈\mu +3/2,}{\sqrt{4st+\lambda 〈2,}}.$  We claim that for $\mu >0$  and $t,s\in R〈+,$  there is a constant $C>0$  such that  $\begin{array}{c}|{\psi }_{\lambda }\left(s,t\right)|⩽Cs〈-\mu -3/2,.\end{array}$ (26)
For $\mu =0$  and $t,s\in R〈+,$  our claim is:  $\begin{array}{c}\left(logt\right)〈2,|{\psi }_{\lambda }\left(s,t\right)|⩽Cmax\left\{1,\left(logs\right)〈2,\right\}s〈-3/2,,\end{array}$ (27)
where $C>0$  . The above claims prove the relations ( 24 ) and ( 25 ). Indeed, consider first the case $\mu >0$  . Then the absolute value of the expression under the integral $I\left(t\right)$  can be estimated by the integrable function $C|e〈-s,-{\sum }_{0⩽j⩽l}\left(-1\right)〈j,s〈j,\frac{}{}j!|s〈-\mu -3/2,\le Cmin\left\{1,s\right\}s〈l,s〈-\mu -3/2,=Cmin\left\{1,s\right\}s〈-\delta -1,,$  where $\delta =\mu +1/2-\left[\mu +1/2\right]$  , $0<\delta <1$  . Then the proof of ( 24 ) is concluded by passing $t\to \infty$  and using the asymptotics of ${w}_{\lambda }$  (see Theorem  3.4 ).
Now, consider the case $\mu =0$  . Observe that the absolute value of the integrand in $\left(logt\right)〈2,I\left(t\right)$  is estimated by the integrable function $Cmax\left\{1,\left(logs\right)〈2,\right\}s〈-3/2,|e〈-s,-1|,so$  ${lim}_{t\to \infty }\left(logt\right)〈2,I\left(t\right)=logx\frac{}{}2\lambda {\int }_{0}〈\infty ,\left(1-e〈-s,\right)s〈-3/2,ds.$  Here we take into account $\left(logt\right)〈2,{\psi }_{\lambda }\left(s,t\right)\approx -logx\frac{}{}2\lambda \left(logt\right)〈2,\frac{}{}s〈3/2,\left(logst\right)〈2,\approx -logx\frac{}{}2\lambda s〈3/2,,t\to \infty ,$  by using the asymptotics of ${w}_{\lambda }$  (see Theorem  3.4 ).
Now we can conclude the proof of the asymptotic behaviour of ${q}_{\mu }\left(t\right)$  in the case $\mu -1/2/\in \mathbb{N}$  .
Note that for $0<\mu <1/2$  we have  $\begin{array}{ccc}t〈\mu +1,{q}_{\mu }\left(t\right)& =& \lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi t}\left(t〈\mu ,x〈\mu -1/2,/2+t〈\mu +1,{\int }_{0}〈\infty ,\left(e〈-\kappa /4t,-1\right){w}_{\lambda }\left(v\right)dv\right)\end{array}$
 $\begin{array}{ccc}& =& \lambda e〈-\lambda 〈2,/4t,\frac{}{}\sqrt{\pi }\left(t〈\mu -1/2,x〈\mu -1/2,/2+I\left(t\right)\right)\to \lambda C/\sqrt{\pi },\end{array}$
where $C$  is the constant from the right-hand side of the formula ( 24 ). The same argument for $\mu =0$  shows that $\left(logt\right)〈2,t{q}_{\mu }\left(t\right)\to \lambda C/\sqrt{\pi },$  where $C$  is the constant from the right-hand side of the formula ( 25 ). For $\mu >1/2$  the asymptotics of ${q}_{\mu }\left(t\right)$  directly follows from the formula ( 24 ).
We now justify our claims ( 26 ) and ( 27 ). We use notation as introduced in the proof of the theorem.
Lemma 4.4. If $\mu >0$  there is $c>0$  such that $|{\psi }_{\lambda }\left(s,t\right)|⩽cs〈-\mu -3/2,.$  For $\mu =0$  there is $C>0$  such that $\left(logt\right)〈2,|{\psi }_{\lambda }\left(s,t\right)|⩽Cmax\left\{1,\left(logs\right)〈2,\right\}s〈-3/2,.$
• Proof. We begin with ( 26 ) first. Let $\mu >0$  . Using boundedness of the function ${w}_{\lambda }$  we can estimate for $ts\le 1$  :
$|{\psi }_{\lambda }\left(s,t\right)|\le {sup}_{v\ge 0}|{w}_{\lambda }\left(v\right)|\frac{t〈\mu +3/2,}{\sqrt{4st+\lambda 〈2,}}\le Cs〈-\mu -3/2,.$  For $ts\ge 1$  we use the asymptotics of ${w}_{\lambda }$  at $\infty$  (see Theorem  3.4 ) to arrive at $|{\psi }_{\lambda }\left(s,t\right)|\le Ct〈\mu +3/2,\left(st\right)〈-\mu -3/2,=Cs〈-\mu -3/2,.$  For $\mu =0$  we estimate our expression as follows. Again using boundedness of the function ${w}_{\lambda }$  we estimate for $ts\le 2$  :
$|{\psi }_{\lambda }\left(s,t\right)|\le {sup}_{v\ge 0}|{w}_{\lambda }\left(v\right)|\frac{t〈3/2,}{\sqrt{4st+\lambda 〈2,}}\le Cs〈-3/2,.$  For $2\le ts$  we use the asymptotics of ${w}_{\lambda }$  to get $\left(logt\right)〈2,|{\psi }_{\lambda }\left(s,t\right)|\le Ct〈3/2,\left(st\right)〈-3/2,\left(logt\right)〈2,\frac{}{}\left(logst\right)〈2,\le \left\{\begin{array}{cc}Cs〈-3/2,\left(logt\right)〈2,,& 2\le ts\le \sqrt{t},\\ Cs〈-3/2,,& ts\ge \sqrt{t}.\end{array}$  Next observe that for $2\le ts\le \sqrt{t}$  we have $\left(logt\right)〈2,\le 4\left(logs\right)〈2,$  .

5 Hiperbolic Brownian motion with drift

Consider the half-space model of the $n$  -dimensional real hyperbolic space $\mathbb{H}〈n,=\left\{\left({x}_{1},\dots ,{x}_{n-1},{x}_{n}\right)\in \mathbb{R}〈n-1,×\mathbb{R}:{x}_{n}>0\right\}.$  The Riemannian metric, the volume element and the Laplace-Beltrami operator are given by $ds〈2,=d{x}_{1}〈2,+...+d{x}_{n-1}〈2,+d{x}_{n}〈2,\frac{}{}{x}_{n}〈2,,$  $dV=d{x}_{1}...d{x}_{n-1}d{x}_{n}\frac{}{}{x}_{n}〈n,,$  $\Delta ={x}_{n}〈2,\left({\sum }_{i=1}〈n,{\partial }_{i}〈2,\right)-\left(n-2\right){x}_{n}{\partial }_{n},$  respectively (here ${\partial }_{i}=\partial \frac{}{}\partial {x}_{i}$  , $i=1,...,n$  ). For $\mu \ge 0$  let $\alpha =2\mu -n+1$  . We also introduce the operator: ${\Delta }_{\mu }=\Delta -\alpha {x}_{n}{\partial }_{n}={x}_{n}〈2,\left({\sum }_{i=1}〈n,{\partial }_{i}〈2,\right)-\left(2\mu -1\right){x}_{n}{\partial }_{n}.$  Let $\left({B}_{i}\left(t\right){\right)}_{i=1...n}$  be a family of independent classical Brownian motions on $\mathbb{R}$  with the generator $d〈2,\frac{}{}dx〈2,$  (and not $1\frac{}{}2d〈2,\frac{}{}dx〈2,$  ) i.e. the variance $E〈0,{B}_{i}〈2,\left(t\right)=2t$  . Then the Brownian motion on $\mathbb{H}〈n,$  , $X=\left({X}_{i}{\right)}_{i=1...n}$  can be described by the following system of stochastic differential equations $\left\{\begin{array}{ccc}d{X}_{1}\left(t\right)& =& {X}_{n}\left(t\right)d{B}_{1}\left(t\right)\\ d{X}_{2}\left(t\right)& =& {X}_{n}\left(t\right)d{B}_{2}\left(t\right)\\ .& .& .\\ d{X}_{n}\left(t\right)& =& {X}_{n}\left(t\right)d{B}_{n}\left(t\right)-\left(n-2\right){X}_{n}\left(t\right)dt.\end{array}$  More generally, if we replace $n-2$  by $\alpha +n-2=2\mu -1$  that the corresponding process will be called the Brownian motion on $\mathbb{H}〈n,$  with drift $\alpha$  . By the Itô formula one verifies that the generator of the solution of this system is ${\Delta }_{\mu }$  . Moreover, it can be easily checked that the solution is given by $\left\{\begin{array}{ccc}{X}_{1}\left(t\right)& =& {X}_{1}\left(0\right)+{\int }_{0}〈t,{X}_{n}\left(t\right)d{B}_{1}\left(s\right)\\ {X}_{2}\left(t\right)& =& {X}_{2}\left(0\right)+{\int }_{0}〈t,{X}_{n}\left(t\right)d{B}_{2}\left(s\right)\\ .& .& .\\ {X}_{n}\left(t\right)& =& {X}_{n}\left(0\right)exp\left({B}_{n}\left(t\right)-2\mu t\right).\end{array}$  Define the projection $\stackrel{~}{:}\mathbb{R}〈n,\ni u=\left({u}_{1},...,{u}_{n}\right)\to \stackrel{~}{u}=\left({u}_{1},...,{u}_{n-1}\right)\in \mathbb{R}〈n-1,$  . In particular, $\stackrel{~}{X}\left(t\right)=\left({X}_{1}\left(t\right),...,{X}_{n-1}\left(t\right)\right)$  . From the representation above one may easily verify ( e.g. by computing Fourier transforms) that
 $\begin{array}{c}\stackrel{~}{X}\left(t\right)\stackrel{d}{=}\stackrel{~}{X}\left(0\right)+\stackrel{~}{B}\left({\int }_{0}〈t,X〈2{,}_{n}\left(s\right)ds\right),t\ge 0.\end{array}$ (28)
where $\stackrel{~}{B}\left(t\right)=\left({B}_{1}\left(t\right),...,{B}_{n-1}\left(t\right)\right)$  is $n-1$  dimensional Brownian motion independent of the process ${X}_{n}\left(t\right)$  . Consider a half-space $D=\left\{u\in \mathbb{H}〈n,:{u}_{n}>a\right\}$  for some fixed $a>0$  . To simplify the notation we choose $a=1$  . Define $\tau =inf\left\{t⩾0:X\left(t\right)/\in D\right\}=inf\left\{t⩾0:{X}_{n}\left(t\right)=1\right\}.$  By ${P}_{1}\left(u,y\right)$  , $u=\left({u}_{1},{u}_{2},...,{u}_{n}\right)\in D$  , $y=\left({y}_{1},{y}_{2},...,{y}_{n-1},1\right)\in \partial D$  we denote the Poisson kernel of $D$  , ie. the density of the distribution of $X\left(\tau \right)$  starting at $u$  (since ${X}_{n}\left(\tau \right)=1$  it is enough to consider the distribution of $\stackrel{~}{X}\left(\tau \right)$  ). From ( 28 ) it is obvious that $\stackrel{~}{X}\left(\tau \right)\stackrel{d}{=}\stackrel{~}{u}+\stackrel{~}{B}\left(A\left(\tau \right)\right),$  where the functional $A\left(\tau \right)$  (starting from ${u}_{n}>1$  ) is independent of $\stackrel{~}{B}\left(t\right)$  . For further considerations we may take $\stackrel{~}{u}=0$  and ${u}_{n}=x>1$  , so the starting point of the $n$  dimensional process $X\left(\cdot \right)$  is $\left(0,...,0,x\right)\in D$  . Since $A\left(\tau \right)$  and $\stackrel{~}{B}\left(t\right)$  are independent, we have the following representation of the Poisson kernel:
Corollary 5.1. ${P}_{1}\left(x,y\right)=1\frac{}{}\left(4\pi \right)〈\left(n-1\right)/2,{\int }_{0}〈\infty ,e〈-|y|〈2,/4t,{q}_{\mu }\left(t\right)dt\frac{}{}t〈\left(n-1\right)/2,.$
Observe that for $\mu =1/2$  the functional $A\left(\tau \right)$  has the standard asymmetric $1/2$  -stable distribution and the Poisson kernel is exactly $n-1$  dimensional Cauchy density, so in what follows, we consider $\mu \ne 1/2$  .
For the sake of simplicity we write $|y|$  as $\rho$  . Taking into account the formula ( 2 ) and ( 2.1 ) we obtain the following representation of the Poisson kernel of the set $D$
Theorem 5.2. For $0⩽\mu <1/2$  we have ${P}_{1}\left(x,y\right)=\Gamma \left(n\frac{}{}2-1\right)\frac{}{}2\pi 〈n/2,\lambda \frac{}{}\left(\lambda 〈2,+\rho 〈2,\right)〈n/2,\left[\left(n-2\right)\left(\lambda +1\right)〈\mu -1/2,-{\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)L〈#,\left(\lambda ,\rho ,v\right)dv\frac{}{}\left(\left(\lambda +v\right)〈2,+\rho 〈2,\right)〈n\frac{}{}2-1,\right],$  where $L〈#,\left(\lambda ,\rho ,v\right)$  is the following function $L〈#,\left(\lambda ,\rho ,v\right)=\left(\lambda 〈2,+\rho 〈2,\right)\left[\left(\left(\lambda +v\right)〈2,+\rho 〈2,\right)〈n/2-1,-\left(\lambda 〈2,+\rho 〈2,\right)〈n/2-1,\right].$  For $\mu >1/2$  we obtain ${P}_{1}\left(x,y\right)=\Gamma \left(n\frac{}{}2-1\right)\frac{}{}2\pi 〈n/2,\lambda \frac{}{}\left(\lambda 〈2,+\rho 〈2,\right)〈n/2,{\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)L\left(\lambda ,\rho ,v\right)dv\frac{}{}\left(\left(\lambda +v\right)〈2,+\rho 〈2,\right)〈n\frac{}{}2-1,,$  with $L\left(\lambda ,\rho ,v\right)$  defined by
 $\begin{array}{ccc}& & L\left(\lambda ,\rho ,v\right)\end{array}$
 $\begin{array}{ccc}& =& \left(n/2-1\right)\left(\left(\lambda +v\right)〈2,-\lambda 〈2,\right)\left(\left(\lambda +v\right)〈2,+\rho 〈2,\right)〈n/2-1,-\left(\lambda 〈2,+\rho 〈2,\right)\left[\left(\left(\lambda +v\right)〈2,+\rho 〈2,\right)〈n/2-1,-\left(\lambda 〈2,+\rho 〈2,\right)〈n/2-1,\right].\end{array}$
• Proof. By the formula ( 5 ) we obtain ${\int }_{0}〈\infty ,e〈-\rho 〈2,/4t,e〈-\lambda 〈2,/4t,dt\frac{}{}t〈1+s,=2〈2s,\Gamma \left(s\right)\frac{}{}\left(\lambda 〈2,+\rho 〈2,\right)〈s,.$  Applying the above formula (with different constants) for three terms appearing in the representation of ${q}_{\mu }$  we obtain, ${\int }_{0}〈\infty ,e〈-\rho 〈2,/4t,e〈-\lambda 〈2,/4t,dt\frac{}{}t〈1+n/2,=2〈n,\Gamma \left(n\frac{}{}2\right)\frac{}{}\left(\lambda 〈2,+\rho 〈2,\right)〈n/2,,$  and  $\begin{array}{ccc}& & {\int }_{0}〈\infty ,e〈-\rho 〈2,/4t,e〈-\lambda 〈2,/4t,{\int }_{0}〈\infty ,e〈-\kappa /4t,{w}_{\lambda }\left(v\right)dt\frac{}{}t〈n/2,\end{array}$
 $\begin{array}{ccc}& =& \Gamma \left(n\frac{}{}2-1\right)2〈n-2,{\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)dv\frac{}{}\left(\left(\lambda +v\right)〈2,+\rho 〈2,\right)〈n\frac{}{}2-1,,\end{array}$
and  $\begin{array}{ccc}& & {\int }_{0}〈\infty ,e〈-\rho 〈2,/4t,e〈-\lambda 〈2,/4t,{\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)dt\frac{}{}t〈n/2,\end{array}$
 $\begin{array}{ccc}& =& \Gamma \left(n\frac{}{}2-1\right)2〈n-2,{\int }_{0}〈\infty ,{w}_{\lambda }\left(v\right)dv\frac{}{}\left(\lambda 〈2,+\rho 〈2,\right)〈n\frac{}{}2-1,.\end{array}$
Taking into account appropriate constants, we obtain the formulas for the Poisson kernel.
Theorem 5.3. ${lim}_{|y|\to \infty }|y|〈n+2\mu -1,{P}_{1}\left(x,y\right)={c}_{\mu },\mu >0;$  ${lim}_{|y|\to \infty }\left(log|y|\right)〈2,|y|〈n-1,{P}_{1}\left(x,y\right)={c}_{0},\mu =0;$  for some positive ${c}_{\mu }$  .
• Proof. From Corollary  5.1 , Theorem  4.1 and arguments presented in its proof (boundedeness of $t〈\mu +1,{q}_{\mu }\left(t\right)$  for $\mu >0$  or $\left(logt\right)〈2,{q}_{\mu }\left(t\right)$  , for $\mu =0$  ) we obtain that the asymptotic behaviour of ${P}_{1}\left(x,y\right)$  is the same (up to a positive constant) as of ${\int }_{0}〈\infty ,t〈-1-\mu -\left(n-1\right)/2,e〈-|y|〈2,/4t,dt=2〈2\mu +n-1,|y|〈-2\mu -n+1,{\int }_{0}〈\infty ,u〈\mu +\left(n-3\right)/2,e〈-u,du,$  for $\mu >0$  . Similar arguments apply for $\mu =0$  .
Acknowledgements The authors would like to thank T. Zak for stimulating conversations on the subject.
References

1. L. Alili, J-C. Gruet, An explanation of generalized Bougerols identity in terms hyperbolic Brownian motion, A collection of research papers. Edited by Marc Yor. Biblioteca de la Revista Matemática Iberoamericana, Madrid, 1997.
2. P. Bougerol, Examples de théorémes locaux sur les groupes résolubles, Ann. Inst. H. Poincare 19 (1983), 369-391.
3. P. Baldi, E. Casadio Tarabusi, A. Figá-Talamanca, Stable laws arising from hitting distributions of processes on homogeneous trees and the hyperbolic half-plane, Pacific J. Math. 197(2)(2001), 257-273.
4. P. Baldi, E. Casadio Tarabusi, A. Figá-Talamanca, M. Yor, Non-symmetric hitting distributions on the hyperbolic half-plane and subordinated perpetuities, Rev. Mat. Iberoam. 17(2001), 587-605.
5. T. Byczkowski, P. Graczyk, A. Stós, Poisson kernels of half–spaces in real hiperbolic spaces, preprint.
6. D. Dufresne, The distribution of a perpetuity, with application to risk theory and pension funding, Scand. Actuarial J., (1990), 39-79.
7. Erdelyi et al., eds., Higher Transcendental Functions, vol. II, McGraw-Hill, New York, 1953-1955.
8. W. Feller, An Introduction to Probability Theory and Its Applications, vol. II, 2d ed. Wiley, New York, 1971.
9. R.K. Getoor, The Brownian Escape Process, Ann. Prob.7(1979), 864-867.
10. R.K. Getoor and M.J. Sharpe, Excursions of Brownian Motion and Bessel Processes, Z.W. 47(1979), 83-106.
11. A.Gonig-Jaeschke and M. Yor, A survey and some generalizations of Bessel process, Bernoulli 9(2)(2003), 313-349.
12. H. Geman, M. Yor, Bessel processes, Asian options and perpetuities, Math. Finance 3(1993), 349-375.
13. M. E. Ismail, Bessel Functions and the infinite divisibility of the Student t-distribution, Ann. Prob.5(1977), 582-585.
14. M. E. Ismail, Integral representations and Complete monotonicity of various quotients of Bessel functions , Canadian J. Math. XXIX(1977), 1198-1207.
15. M. E. Ismail and D.H. Kelker, Special Functions, Stiltjes Transforms, and Infinite divisibility, Siam J. Math. Anal. 10 (1976), 884-901.
16. J. Kent, Some Probabilistic Properties of Bessel Functions, Ann. Prob.6(1978), 760-770.
17. J. Lamperti, Semi-Stable Markov Processes I, Z.W. 22(1972), 205-225.
18. C. Monthus, Étude de quelques fonctionnelles du movement Brownien et de certaines propriétés de la diffusion unidimensinelle en milieu aleatoire, Thése de Doctorat de lUniversité Paris VI, 1995.
19. J. Pitman and M. Yor, Bessel processes and infinitely divisible laws, LNM 851: Stochastic Integrals, Springer (1981).
20. F. Spitzer, Some theorems concerning 2-dimensional Brownian motion, Trans. Amer. Math. Soc. 87(1958), 187-197.
21. D. Revuz and M. Yor, Continuous martingales and Brownian motion, 2d ed. Springer, 1994.
22. M. Yor, Sur certaines fonctionnelles du mouvement Brownien réel, J. Appl. Prob. 29(1992), 202-208.
23. M. Yor, On some exponential functionals of Brownian Motion, Adv. Appl. Prob. 24(1992), 509-531.
24. Exponential functionals and principal values related to Brownian motion. A collection of research papers. Edited by Marc Yor. Biblioteca de la Revista Matemática Iberoamericana, Madrid, 1997.