getdoc32e8. 122KB Jun 04 2011 12:04:14 AM
Elect. Comm. in Probab. 7 (2002) 37–46
ELECTRONIC
COMMUNICATIONS
in PROBABILITY
FURTHER EXPONENTIAL GENERALIZATION
OF PITMAN’S 2M-X THEOREM
FABRICE BAUDOIN
Laboratoire de Probabilit´es et Mod`eles al´eatoires,
Universit´es Paris 6 et Paris 7, Laboratoire de Finance-Assurance CREST
email: baudoin@ensae.fr
submitted November 5, 2001 Final version accepted January 11, 2002
AMS 2000 Subject classification: 60J60
Diffusion processes, Exponential analogue of the 2M − X Pitman’s theorem
Abstract
We present a class of processes which enjoy an exponential analogue of Pitman’s 2M-X theorem,
improving hence some works of H. Matsumoto and M. Yor.
1
Introduction
In a recent series of papers [6-13], H. Matsumoto and M. Yor have shown the following theorem:
Theorem 1 Let et := exp(Bt + µt), t ≥ 0, a geometric Brownian motion with drift µ ∈ R∗+ .
The following identity (called Dufresne identity) holds
(µ)
where
R +∞
0
Rt
1
(−µ) 2
(e
) ds
0 s
(−µ) 2
(e
es
law
, t > 0 = Rt
) ds is a copy of
Moreover, the process
Rt
0
2
(e(µ)
s ) ds
(µ)
et
1
(µ)
(e )2 ds
0 s
R +∞
0
(−µ) 2
(es
+ R +∞
0
1
(−µ)
(e
es )2 ds
,t>0
) ds independent of the process
R
t (µ) 2
0 (es ) ds t≥0
.
, t ≥ 0, is, in its own filtration which is strictly included in
the original Brownian filtration, a diffusion with infinitesimal generator
1
K1+µ 1 d
1 2 d2
+ −µ+ x+
G= x
2 dx2
2
Kµ x dx
As shown in [9], by an elementary Laplace method argument, this theorem allows to recover
the classical 2M − X Pitman’s theorem.
In this work, we show that theorem 1 can be extended to a large class of processes which
are constructed from the geometric Brownian motion by a simple transformation and which
satisfy a generalized Dufresne identity (originally discussed in [5] and [10]).
This paper is essentially self-contained and contains a new proof of theorem 1.
37
38
Electronic Communications in Probability
2
Diffusions for which (Xt−1
Rt
0
Xs2 ds)t≥0 is also a diffusion
Let (Ω, (Ft )t≥0 , P) a filtered probability space on which a standard Brownian motion (Bt )t≥0
and Z a positive F∞ -measurable variable independent of the process (Bt )t≥0 are defined.
In [6], C. Donati-Martin, H. Matsumoto and M. Yor introduced the following transforms from
C (R+ , R) to itself
Z t
exp (2φ (s)) ds
Tα (φ)t = φ (t) − ln 1 + α
0
In their paper these authors have shown that if the law of Z is equivalent to the Lebesgue
(µ)
measure, then the laws of the processes TZ B (µ) s , s ≤ t, and Bs = Bs + µs, s ≤ t are
equivalent.
In what follows we study TZ B (µ) in its own filtration and we characterize the variables Z
such that TZ B (µ) is a diffusion. As a consequence of this characterization, we will be able
to give a new proof of Matsumoto-Yor’s theorem 1. More precisely, in all this paper, we deal
with the process (Xt )t≥0 defined as follows
(µ)
Xt =
et
Rt
1+Z
0
(µ)
(es )2 ds
,t≥0
where et := x0 exp(Bt + µt), t ≥ 0, is the geometric Brownian motion with drift µ ∈ R∗+ and
started at x0 > 0.
(µ)
Rt
Proposition 2 The process ( 0 Xs2 ds)t≥0 converges
following generalized Dufresne identity holds:
Moreover,
Rt
0
1
Xs2 ds
Rt
0
, t>0
Xt
Xs2 ds
s
a.s.
,t>0
t
, t > 0) is independent of
thus ( R t X
X 2 ds
0
=
a.s.
=
R +∞
0
Rt
0
P
a.s. when t → +∞ to
1
(µ)
(es )2 ds
+Z , t>0
(µ)
Rt
et
(µ) 2
0 (es ) ds
1
Z
and the
, t>0 ,
Xs2 ds.
Theorem 3 Assume that the law of Z admits a bounded C 2 density with respect to the law
2
γ . Then (Xt )t≥0 is a diffusion in its own filtration if and only if there exists δ ≥ 0 such
x20 µ
that:
x2
δ2
xµ
0
xµ−1 e− 2x − 2 x dx , x > 0
P (Z ∈ dx) = µ 0
(2.1)
2δ Kµ (δx0 )
Moreover, in this case there exists a standard Brownian motion (βt )t≥0 adapted to the natural
filtration of (Xt )t≥0 such that
K1+µ (δXt )
1
dt + dβt , t ≥ 0
dXt = Xt µ + − δXt
2
Kµ (δXt )
(2.2)
Pitman’s 2M − X Theorem
39
law 2
γ
x20 µ
Remark 4 For δ = 0, Z =
and (2.2) becomes
dXt = Xt
Rt
h
−µ+
X 2 ds
i
1
dt + dβt , t ≥ 0
2
Corollary 5 The process Zt := 0 Xts , t ≥ 0, is, in its own filtration, a diffusion independent
R +∞
of 0 Xs2 ds with infinitesimal generator
G x0 =
h
1 2 d2
K1+µ x0 i d
1
x
+
x
x
+
−
µ
+
0
2 dx2
2
Kµ x dx
Moreover, under the assumption (2.1) , for t > 0 and x, z > 0
P (Xt ∈ dx | Zt , Zt = z) =
x 2µ K (δx)
1
δ2
µ
e− 2 zx− 2z
x0
Kµ (δx0 )
where (Zt )t≥0 is the natural filtration of (Zt )t≥0 = (
Rt
0
x
x0
+
x0
x
dx
2xKµ
Xs2 ds
)t≥0 .
Xt
1
z
Proposition 6 Under the assumption (2.1), for all t ≥ 0
Z +∞
X2
− δ2
E R +∞ t
Xs2 ds Xt = 2µ
2 ds
X
t
s
t
(2.3)
where (Xt )t≥0 is the natural filtration of (Xt )t≥0 .
In the following remarks, we assume that (2.1) is satisfied for some δ ≥ 0.
Remark 7
1. We recall the definition of the Mac-Donald function
1 x µ
Kµ (x) =
2 2
Z
+∞
0
x2
e−t− 4t
dt
t1+µ
but for further details, we refer to [7].
2. The law (2.1) is called a generalized inverse Gaussian distribution. These laws have been
widely discussed by Barndorff-Nielsen [1], Letac-Wesolowski [8], Matsumoto-Yor [12],
and Vallois [19] among others.
3. We get the theorem 1.1. by taking δ = 0, because in this case
law
(−µ)
(Xt , t ≥ 0) = (et
, t ≥ 0)
4. The stochastic differential equation (2.2) solved by (Xt )t≥0 , enjoys the pathwise uniqueness property. Indeed, a simple computation shows that a scale function is given by
s(x) =
Iµ
(δx)
Kµ
40
Electronic Communications in Probability
where Iµ is the classical modified Bessel function with index µ. Using Wronskian relation
for Bessel functions, we get then the speed measure
2
2
m (dx) = − Kµ (δx) dx
x
And finally, Feller’s test for explosions (see for example [17] pp.386) gives
P (e = +∞) = 1
where e is the lifetime of a solution.
5. If we apply Itˆ
o’s formula conditionally to Z, we see that in the natural filtration of (Xt )t≥0
R +∞ 2
enlarged by 0 Xs ds the process (Xt )t≥0 is a semimartingale whose decomposition is
!
#
"
Xt2
1
dt + dBt , t ≥ 0
µ + − R +∞
dXt = Xt
2
Xs2 ds
t
6. For µ = 21 , (2.2) becomes
dXt = −δXt2 dt + Xt dβt
and this equation is solved by
1
Xt =
x0 e− 2 t+βt
, t≥0
Rt 1
1 + δx0 0 e− 2 s+βs ds
We have then the following surprising identity in law
!
!
1
1
x0 e 2 t+Bt
x0 e− 2 t+βt
law
, t≥0
, t≥0 =
Rt
Rt 1
1 + δx0 0 e− 2 s+βs ds
1 + Zx20 0 es+2Bs ds
for which it would be interesting to have a direct proof.
For µ 6= 21 , it seems difficult to solve explicitly the equation (2.2) , even in the case
µ = n + 12 , with n ∈ N . Let us just mention that
Xt =
solves
dXt =
x0 e−µt+βt
Rt
1 + δx0 0 e−µs+βs ds
1
2
− µ + Xt − δXt dt + Xt dβt
2
which coincides with (2.2) if and only if µ = 12 .
7. The process (ρt )t≥0 related to (Xt )t≥0 by the Lamperti’s relation Xt = ρR t Xs2 ds , t ≥ 0,
0
is a well-known diffusion discussed in [11], [15], [16], and [20] the BES (µ, δ ↓) process,
i.e. the diffusion process with the infinitesimal generator
µ + 12
K1+µ (δx) d
1 d2
−
δ
+
2 dx2
x
Kµ (δx)
dx
From [6] (see also [3]), it implies then that there exists a Bessel process Q with index µ,
starting at x0 and independent of the first hitting T0 of 0 for ρ such that
t
law
1−
Q T0 t , t < T0
(ρt , t < T0 ) =
T0 −t
T0
Pitman’s 2M − X Theorem
3
41
Proofs
Proof of proposition 2
This proof is very simple and inspired from [6].
As
(µ)
et
,t≥0
Xt =
R t (µ)
1 + Z 0 (es )2 ds
we have
Xt2 =
and hence
Z
t
0
(µ)
et
Rt
(µ) 2
0 (es ) ds
1+Z
Xs2 ds
=
Rt
2
(3.1)
,t≥0
(µ)
(es )2 ds
,t≥0
R t (µ)
1 + Z 0 (es )2 ds
0
(3.2)
This equality implies immediately the P−a.s. convergence of the process
1
Z , it also implies
1
1
= R t (µ)
+Z , t >0
Rt
2
2 ds
(e
)
s
0 Xs ds
0
Rt
0
Xs2 ds, t ≥ 0, to
Finally, by dividing (3.2) by (3.1), we deduce
Rt
0
Xs2 ds
=
Xt
Rt
0
Proof of theorem 3
Let us set in all the proof
1
=
Y =
Z
(µ)
(es )2 ds
,t≥0
(µ)
et
Z
+∞
0
Xs2 ds
Let now y > 0 and denote Q y the law of the process (Xt )t≥0 conditioned with Y = y.
From theorem 1.5. of [6], we see that the following absolute continuity relation takes place
χ
1+µ x0 −
R tt
χt 2µ
y
2
=
e 2y 2(y− 0 χs ds) 1R t χ2s ds 0
2 Γ (µ) x1+µ
where ξ is the bounded density of Y with respect to x22γµ , we deduce after some elementary
0
computation that the law Q of our process (Xt )t≥0 satisfies the following equivalence relation
dQ /Ξt =
Z
0
+∞
e−u uµ−1
ξ
Γ (µ)
Z
0
t
χ2s ds +
χ2t
2u
du
dP−µ
/Ξt , t ≥ 0
(3.3)
42
Electronic Communications in Probability
Hence, by Girsanov theorem, if X is a diffusion in its own filtration, then the infinitesimal
generator of this diffusion can be written
d
1
1 2 d
+ x −µ + + xb (x)
L= x
2
2 dx
2
dx
where b : R∗+ → R is related to
ϕ (t, x) =
Z
+∞
0
x2
e−u uµ−1
ξ t+
du
Γ (µ)
2u
by the Hopf-Cole transformation
∂
ln ϕ
(3.4)
∂x
From (3.4) the homogeneity of b implies that there exist two functions f and g such that
b=
ϕ (t, x) = f (x) g (t) , t, x > 0
but ϕ is a solution of the following partial differential equation
∂ϕ 1 ∂ 2 ϕ −µ +
+
+
∂t
2 ∂x2
x
1
2
∂ϕ
=0
∂x
it immediately implies that there exists a constant C such that
g (t) = eCt
We note now that the constant C is negative, because ξ is bounded and we have the limit
condition
(3.5)
lim ϕ (t, x) = ξ (t)
x→0+
Denote the constant C by − δ2 with δ ∈ R+ and conclude that
2
δ2
x2
0
e− 2 x− 2x
xµ
dx , x > 0
P (Y ∈ dx) = µ 0
2δ Kµ (δx0 ) x1+µ
It proves the first part of our theorem.
On the other hand, if (2.1) is satisfied the following equivalence relation takes place (it suffices
to use the equivalence relation (3.3))
µ
2 R
Kµ (δχt )
χt
− δ2 0t χ2s ds
dP−µ
(3.6)
dQ /Ξt = e
/Ξt , t ≥ 0
x0
Kµ (δx0 )
which implies, by Girsanov theorem
K1+µ (δXt )
1
dt + dβt , t ≥ 0
dXt = Xt µ + − δXt
2
Kµ (δXt )
where (βt )t≥0 is a
P standard Brownian motion adapted to the natural filtration of (Xt )t≥0 .
Proof of corollary 5
Pitman’s 2M − X Theorem
43
Let δ > 0.
Let us consider a sequence (Zn )n>0 of random variables independent of (Bt )t≥0 and such that
P (Zn ∈ dx) =
2
δ2
nµ
µ−1 − 2x
− n2 x
x
e
dx , x > 0
2δ µ Kµ (δn)
(n)
We associate with this sequence (Zn )n>0 the sequence of processes Xt
(n)
Xt
(n)
Xt
defined by
n exp (Bt + µt)
,t≥0
Rt
1 + Zn n2 0 exp (2Bs + 2µs) ds
=
We have
1
t≥0
Rt
1
= exp (−Bt − µt) + Zn n
n
0
exp (2Bs + 2µs) ds
exp (Bt + µt)
But, one can show that
law
nZn →n→+∞ δ
hence
1
(n)
Xt
,t≥0
!
→law
n→+∞
On the other hand, from theorem 3,
Gδ =
1 2 d2
x
+
2 dx2
1
(n)
Xt
!
exp (2Bs + 2µs) ds
,t≥0
δ
exp (Bt + µt)
, t ≥ 0 is a diffusion with infinitesimal generator
−µ +
Rt
0
1
2
x+δ
K1+µ
Kµ
d
δ
x
dx
Now, to conclude the proof of the first part of the corollary, we use proposition 2 which asserts
that
R t (µ) 2
Rt 2
es
ds
X
ds
0
s
0
,t≥0
=
(µ)
Xt
et
The computation of the conditional probabilities is easily deduced from (3.6) and proposition
6 of [9], by a change of probability, namely the absolute continuity (3.6).
Proof of proposition 6
A simple computation shows that for t ≥ 0
Dt
Z
0
+∞
e2χs −2µs ds = 2
Z
+∞
e2χs −2µs ds
t
where D is the Malliavin’s differential and (χt )t≥0 the coordinate process on the Wiener space.
2 R +∞
Clark-Ocone’s formula (see [4]) applied to exp{− δ2 0 e2χs −2µs ds} implies then that for all
t≥0
Z +∞
Xt K1+µ (δXt ) 2µ
− 2
E
(3.7)
Xs2 ds | Xt =
δ Kµ (δXt )
δ
t
44
Electronic Communications in Probability
R
+∞
As noticed in the remarks of section 2, in the enlarged filtration Xt ∨ σ 0 Xs2 ds
t≥0
,
our process (Xt )t≥0 is a semimartingale whose decomposition is
!
#
"
Xt2
1
dt + dBt , t ≥ 0
µ + − R +∞
dXt = Xt
2
Xs2 ds
t
By projecting this decomposition on (Xt )t≥0 , using the classical filtering formulas, we get
1
δ K1+µ (δXt )
E R +∞
Xt =
2
X
t Kµ (δXt )
Xs ds
t
(3.8)
And finally, (3.7) and (3.8) imply (2.3)
4
Opening
To conclude the paper, let us relate our result to another simple transformation of the Brownian
motion (first considered by T. Jeulin and M. Yor and then generalized by P.A. Meyer, see [14]).
It is easily seen that for a standard Brownian motion (Bt )0≤t
ELECTRONIC
COMMUNICATIONS
in PROBABILITY
FURTHER EXPONENTIAL GENERALIZATION
OF PITMAN’S 2M-X THEOREM
FABRICE BAUDOIN
Laboratoire de Probabilit´es et Mod`eles al´eatoires,
Universit´es Paris 6 et Paris 7, Laboratoire de Finance-Assurance CREST
email: baudoin@ensae.fr
submitted November 5, 2001 Final version accepted January 11, 2002
AMS 2000 Subject classification: 60J60
Diffusion processes, Exponential analogue of the 2M − X Pitman’s theorem
Abstract
We present a class of processes which enjoy an exponential analogue of Pitman’s 2M-X theorem,
improving hence some works of H. Matsumoto and M. Yor.
1
Introduction
In a recent series of papers [6-13], H. Matsumoto and M. Yor have shown the following theorem:
Theorem 1 Let et := exp(Bt + µt), t ≥ 0, a geometric Brownian motion with drift µ ∈ R∗+ .
The following identity (called Dufresne identity) holds
(µ)
where
R +∞
0
Rt
1
(−µ) 2
(e
) ds
0 s
(−µ) 2
(e
es
law
, t > 0 = Rt
) ds is a copy of
Moreover, the process
Rt
0
2
(e(µ)
s ) ds
(µ)
et
1
(µ)
(e )2 ds
0 s
R +∞
0
(−µ) 2
(es
+ R +∞
0
1
(−µ)
(e
es )2 ds
,t>0
) ds independent of the process
R
t (µ) 2
0 (es ) ds t≥0
.
, t ≥ 0, is, in its own filtration which is strictly included in
the original Brownian filtration, a diffusion with infinitesimal generator
1
K1+µ 1 d
1 2 d2
+ −µ+ x+
G= x
2 dx2
2
Kµ x dx
As shown in [9], by an elementary Laplace method argument, this theorem allows to recover
the classical 2M − X Pitman’s theorem.
In this work, we show that theorem 1 can be extended to a large class of processes which
are constructed from the geometric Brownian motion by a simple transformation and which
satisfy a generalized Dufresne identity (originally discussed in [5] and [10]).
This paper is essentially self-contained and contains a new proof of theorem 1.
37
38
Electronic Communications in Probability
2
Diffusions for which (Xt−1
Rt
0
Xs2 ds)t≥0 is also a diffusion
Let (Ω, (Ft )t≥0 , P) a filtered probability space on which a standard Brownian motion (Bt )t≥0
and Z a positive F∞ -measurable variable independent of the process (Bt )t≥0 are defined.
In [6], C. Donati-Martin, H. Matsumoto and M. Yor introduced the following transforms from
C (R+ , R) to itself
Z t
exp (2φ (s)) ds
Tα (φ)t = φ (t) − ln 1 + α
0
In their paper these authors have shown that if the law of Z is equivalent to the Lebesgue
(µ)
measure, then the laws of the processes TZ B (µ) s , s ≤ t, and Bs = Bs + µs, s ≤ t are
equivalent.
In what follows we study TZ B (µ) in its own filtration and we characterize the variables Z
such that TZ B (µ) is a diffusion. As a consequence of this characterization, we will be able
to give a new proof of Matsumoto-Yor’s theorem 1. More precisely, in all this paper, we deal
with the process (Xt )t≥0 defined as follows
(µ)
Xt =
et
Rt
1+Z
0
(µ)
(es )2 ds
,t≥0
where et := x0 exp(Bt + µt), t ≥ 0, is the geometric Brownian motion with drift µ ∈ R∗+ and
started at x0 > 0.
(µ)
Rt
Proposition 2 The process ( 0 Xs2 ds)t≥0 converges
following generalized Dufresne identity holds:
Moreover,
Rt
0
1
Xs2 ds
Rt
0
, t>0
Xt
Xs2 ds
s
a.s.
,t>0
t
, t > 0) is independent of
thus ( R t X
X 2 ds
0
=
a.s.
=
R +∞
0
Rt
0
P
a.s. when t → +∞ to
1
(µ)
(es )2 ds
+Z , t>0
(µ)
Rt
et
(µ) 2
0 (es ) ds
1
Z
and the
, t>0 ,
Xs2 ds.
Theorem 3 Assume that the law of Z admits a bounded C 2 density with respect to the law
2
γ . Then (Xt )t≥0 is a diffusion in its own filtration if and only if there exists δ ≥ 0 such
x20 µ
that:
x2
δ2
xµ
0
xµ−1 e− 2x − 2 x dx , x > 0
P (Z ∈ dx) = µ 0
(2.1)
2δ Kµ (δx0 )
Moreover, in this case there exists a standard Brownian motion (βt )t≥0 adapted to the natural
filtration of (Xt )t≥0 such that
K1+µ (δXt )
1
dt + dβt , t ≥ 0
dXt = Xt µ + − δXt
2
Kµ (δXt )
(2.2)
Pitman’s 2M − X Theorem
39
law 2
γ
x20 µ
Remark 4 For δ = 0, Z =
and (2.2) becomes
dXt = Xt
Rt
h
−µ+
X 2 ds
i
1
dt + dβt , t ≥ 0
2
Corollary 5 The process Zt := 0 Xts , t ≥ 0, is, in its own filtration, a diffusion independent
R +∞
of 0 Xs2 ds with infinitesimal generator
G x0 =
h
1 2 d2
K1+µ x0 i d
1
x
+
x
x
+
−
µ
+
0
2 dx2
2
Kµ x dx
Moreover, under the assumption (2.1) , for t > 0 and x, z > 0
P (Xt ∈ dx | Zt , Zt = z) =
x 2µ K (δx)
1
δ2
µ
e− 2 zx− 2z
x0
Kµ (δx0 )
where (Zt )t≥0 is the natural filtration of (Zt )t≥0 = (
Rt
0
x
x0
+
x0
x
dx
2xKµ
Xs2 ds
)t≥0 .
Xt
1
z
Proposition 6 Under the assumption (2.1), for all t ≥ 0
Z +∞
X2
− δ2
E R +∞ t
Xs2 ds Xt = 2µ
2 ds
X
t
s
t
(2.3)
where (Xt )t≥0 is the natural filtration of (Xt )t≥0 .
In the following remarks, we assume that (2.1) is satisfied for some δ ≥ 0.
Remark 7
1. We recall the definition of the Mac-Donald function
1 x µ
Kµ (x) =
2 2
Z
+∞
0
x2
e−t− 4t
dt
t1+µ
but for further details, we refer to [7].
2. The law (2.1) is called a generalized inverse Gaussian distribution. These laws have been
widely discussed by Barndorff-Nielsen [1], Letac-Wesolowski [8], Matsumoto-Yor [12],
and Vallois [19] among others.
3. We get the theorem 1.1. by taking δ = 0, because in this case
law
(−µ)
(Xt , t ≥ 0) = (et
, t ≥ 0)
4. The stochastic differential equation (2.2) solved by (Xt )t≥0 , enjoys the pathwise uniqueness property. Indeed, a simple computation shows that a scale function is given by
s(x) =
Iµ
(δx)
Kµ
40
Electronic Communications in Probability
where Iµ is the classical modified Bessel function with index µ. Using Wronskian relation
for Bessel functions, we get then the speed measure
2
2
m (dx) = − Kµ (δx) dx
x
And finally, Feller’s test for explosions (see for example [17] pp.386) gives
P (e = +∞) = 1
where e is the lifetime of a solution.
5. If we apply Itˆ
o’s formula conditionally to Z, we see that in the natural filtration of (Xt )t≥0
R +∞ 2
enlarged by 0 Xs ds the process (Xt )t≥0 is a semimartingale whose decomposition is
!
#
"
Xt2
1
dt + dBt , t ≥ 0
µ + − R +∞
dXt = Xt
2
Xs2 ds
t
6. For µ = 21 , (2.2) becomes
dXt = −δXt2 dt + Xt dβt
and this equation is solved by
1
Xt =
x0 e− 2 t+βt
, t≥0
Rt 1
1 + δx0 0 e− 2 s+βs ds
We have then the following surprising identity in law
!
!
1
1
x0 e 2 t+Bt
x0 e− 2 t+βt
law
, t≥0
, t≥0 =
Rt
Rt 1
1 + δx0 0 e− 2 s+βs ds
1 + Zx20 0 es+2Bs ds
for which it would be interesting to have a direct proof.
For µ 6= 21 , it seems difficult to solve explicitly the equation (2.2) , even in the case
µ = n + 12 , with n ∈ N . Let us just mention that
Xt =
solves
dXt =
x0 e−µt+βt
Rt
1 + δx0 0 e−µs+βs ds
1
2
− µ + Xt − δXt dt + Xt dβt
2
which coincides with (2.2) if and only if µ = 12 .
7. The process (ρt )t≥0 related to (Xt )t≥0 by the Lamperti’s relation Xt = ρR t Xs2 ds , t ≥ 0,
0
is a well-known diffusion discussed in [11], [15], [16], and [20] the BES (µ, δ ↓) process,
i.e. the diffusion process with the infinitesimal generator
µ + 12
K1+µ (δx) d
1 d2
−
δ
+
2 dx2
x
Kµ (δx)
dx
From [6] (see also [3]), it implies then that there exists a Bessel process Q with index µ,
starting at x0 and independent of the first hitting T0 of 0 for ρ such that
t
law
1−
Q T0 t , t < T0
(ρt , t < T0 ) =
T0 −t
T0
Pitman’s 2M − X Theorem
3
41
Proofs
Proof of proposition 2
This proof is very simple and inspired from [6].
As
(µ)
et
,t≥0
Xt =
R t (µ)
1 + Z 0 (es )2 ds
we have
Xt2 =
and hence
Z
t
0
(µ)
et
Rt
(µ) 2
0 (es ) ds
1+Z
Xs2 ds
=
Rt
2
(3.1)
,t≥0
(µ)
(es )2 ds
,t≥0
R t (µ)
1 + Z 0 (es )2 ds
0
(3.2)
This equality implies immediately the P−a.s. convergence of the process
1
Z , it also implies
1
1
= R t (µ)
+Z , t >0
Rt
2
2 ds
(e
)
s
0 Xs ds
0
Rt
0
Xs2 ds, t ≥ 0, to
Finally, by dividing (3.2) by (3.1), we deduce
Rt
0
Xs2 ds
=
Xt
Rt
0
Proof of theorem 3
Let us set in all the proof
1
=
Y =
Z
(µ)
(es )2 ds
,t≥0
(µ)
et
Z
+∞
0
Xs2 ds
Let now y > 0 and denote Q y the law of the process (Xt )t≥0 conditioned with Y = y.
From theorem 1.5. of [6], we see that the following absolute continuity relation takes place
χ
1+µ x0 −
R tt
χt 2µ
y
2
=
e 2y 2(y− 0 χs ds) 1R t χ2s ds 0
2 Γ (µ) x1+µ
where ξ is the bounded density of Y with respect to x22γµ , we deduce after some elementary
0
computation that the law Q of our process (Xt )t≥0 satisfies the following equivalence relation
dQ /Ξt =
Z
0
+∞
e−u uµ−1
ξ
Γ (µ)
Z
0
t
χ2s ds +
χ2t
2u
du
dP−µ
/Ξt , t ≥ 0
(3.3)
42
Electronic Communications in Probability
Hence, by Girsanov theorem, if X is a diffusion in its own filtration, then the infinitesimal
generator of this diffusion can be written
d
1
1 2 d
+ x −µ + + xb (x)
L= x
2
2 dx
2
dx
where b : R∗+ → R is related to
ϕ (t, x) =
Z
+∞
0
x2
e−u uµ−1
ξ t+
du
Γ (µ)
2u
by the Hopf-Cole transformation
∂
ln ϕ
(3.4)
∂x
From (3.4) the homogeneity of b implies that there exist two functions f and g such that
b=
ϕ (t, x) = f (x) g (t) , t, x > 0
but ϕ is a solution of the following partial differential equation
∂ϕ 1 ∂ 2 ϕ −µ +
+
+
∂t
2 ∂x2
x
1
2
∂ϕ
=0
∂x
it immediately implies that there exists a constant C such that
g (t) = eCt
We note now that the constant C is negative, because ξ is bounded and we have the limit
condition
(3.5)
lim ϕ (t, x) = ξ (t)
x→0+
Denote the constant C by − δ2 with δ ∈ R+ and conclude that
2
δ2
x2
0
e− 2 x− 2x
xµ
dx , x > 0
P (Y ∈ dx) = µ 0
2δ Kµ (δx0 ) x1+µ
It proves the first part of our theorem.
On the other hand, if (2.1) is satisfied the following equivalence relation takes place (it suffices
to use the equivalence relation (3.3))
µ
2 R
Kµ (δχt )
χt
− δ2 0t χ2s ds
dP−µ
(3.6)
dQ /Ξt = e
/Ξt , t ≥ 0
x0
Kµ (δx0 )
which implies, by Girsanov theorem
K1+µ (δXt )
1
dt + dβt , t ≥ 0
dXt = Xt µ + − δXt
2
Kµ (δXt )
where (βt )t≥0 is a
P standard Brownian motion adapted to the natural filtration of (Xt )t≥0 .
Proof of corollary 5
Pitman’s 2M − X Theorem
43
Let δ > 0.
Let us consider a sequence (Zn )n>0 of random variables independent of (Bt )t≥0 and such that
P (Zn ∈ dx) =
2
δ2
nµ
µ−1 − 2x
− n2 x
x
e
dx , x > 0
2δ µ Kµ (δn)
(n)
We associate with this sequence (Zn )n>0 the sequence of processes Xt
(n)
Xt
(n)
Xt
defined by
n exp (Bt + µt)
,t≥0
Rt
1 + Zn n2 0 exp (2Bs + 2µs) ds
=
We have
1
t≥0
Rt
1
= exp (−Bt − µt) + Zn n
n
0
exp (2Bs + 2µs) ds
exp (Bt + µt)
But, one can show that
law
nZn →n→+∞ δ
hence
1
(n)
Xt
,t≥0
!
→law
n→+∞
On the other hand, from theorem 3,
Gδ =
1 2 d2
x
+
2 dx2
1
(n)
Xt
!
exp (2Bs + 2µs) ds
,t≥0
δ
exp (Bt + µt)
, t ≥ 0 is a diffusion with infinitesimal generator
−µ +
Rt
0
1
2
x+δ
K1+µ
Kµ
d
δ
x
dx
Now, to conclude the proof of the first part of the corollary, we use proposition 2 which asserts
that
R t (µ) 2
Rt 2
es
ds
X
ds
0
s
0
,t≥0
=
(µ)
Xt
et
The computation of the conditional probabilities is easily deduced from (3.6) and proposition
6 of [9], by a change of probability, namely the absolute continuity (3.6).
Proof of proposition 6
A simple computation shows that for t ≥ 0
Dt
Z
0
+∞
e2χs −2µs ds = 2
Z
+∞
e2χs −2µs ds
t
where D is the Malliavin’s differential and (χt )t≥0 the coordinate process on the Wiener space.
2 R +∞
Clark-Ocone’s formula (see [4]) applied to exp{− δ2 0 e2χs −2µs ds} implies then that for all
t≥0
Z +∞
Xt K1+µ (δXt ) 2µ
− 2
E
(3.7)
Xs2 ds | Xt =
δ Kµ (δXt )
δ
t
44
Electronic Communications in Probability
R
+∞
As noticed in the remarks of section 2, in the enlarged filtration Xt ∨ σ 0 Xs2 ds
t≥0
,
our process (Xt )t≥0 is a semimartingale whose decomposition is
!
#
"
Xt2
1
dt + dBt , t ≥ 0
µ + − R +∞
dXt = Xt
2
Xs2 ds
t
By projecting this decomposition on (Xt )t≥0 , using the classical filtering formulas, we get
1
δ K1+µ (δXt )
E R +∞
Xt =
2
X
t Kµ (δXt )
Xs ds
t
(3.8)
And finally, (3.7) and (3.8) imply (2.3)
4
Opening
To conclude the paper, let us relate our result to another simple transformation of the Brownian
motion (first considered by T. Jeulin and M. Yor and then generalized by P.A. Meyer, see [14]).
It is easily seen that for a standard Brownian motion (Bt )0≤t