Probabilites

Moment identities for Poisson-Skorohod integrals

and application to measure invariance
Nicolas Privault
Abstract -

We present a moment identity on the Poisson space that extends the Skorohod
isometry to arbitrary powers of the Skorohod integral. Applications of this identity are given
to the invariance of Poisson measures under intensity preserving random transformations.

Identit
es de moments pour les int
egrales de Poisson-Skorohod et
applications `
a linvariance en mesure
R
esum
e - Nous presentons une identite de moments sur lespace de Poisson qui etend
lisometrie de Skorohod `
a des puissances quelconques de lintegrale de Skorohod, et nous
etudions les applications de cette identite `a linvariance de la mesure de Poisson sous les tranformations aleatoires qui preservent lintensite.
Key words: Poisson measures, random transformations, invariance, Skorohod integral, moment identities.
Mathematics Subject Classification: 60G57, 60G30, 60H07, 28D05, 28C20.

Introduction

The classical invariance theorem for Poisson measures states that given a deterministic
transformation : X Y between measure spaces (X, ) and (Y, ) sending to ,
the corresponding transformation on point processes maps the Poisson distribution
with intensity (dx) on X to the Poisson distribution with intensity (dy) on Y . In
this note we present sufficient conditions for the invariance of random transformations
: X X Y of Poisson random measures on metric spaces. Our results are inspired
by the treatment of the Wiener case in [8], see [6] for a recent simplified proof. However,
the use of finite difference operators instead of derivation operators as in the continuous
case makes the proofs and arguments more complex from an algebraic point of view.
Here the almost sure isometry condition on Rd assumed in the Gaussian case will be
replaced by an almost sure condition on the preservation of intensity measures and, as
in the Wiener case, we will characterize probability measures via their moments.
In this Note the proofs of the main results are only outlined. The details of the complete
proofs, which are technical, can be found in [5].

Notation and preliminaries

In this section we recall some notation and facts on stochastic analysis under Poisson
measures, see [3] and [7] for recent reviews. Let X be a -compact metric space with
1

Borel -algebra B(X). Let X denote the configuration space on X, i.e. the space of at
most countable and locally finite subsets of X, defined as


X = = (xi )N
i=1 X, xi 6= xj i 6= j, N N {} ,
and endowed with the Poisson probability measure with -finite diffuse intensity (dx)
(X)
X
X
on X. Each element of is identified to the Radon point measure =
xi , where
i=1

x denotes the Dirac measure at x X and (X) N {} is the cardinality of . Let

D denote the finite difference gradient defined (d, dx)-almost everywhere as
Dx F () = F ( {x}) F (),

X ,

x X,

(2.1)

for any random variable F : X R, cf. [1]. We refer to [2] for the definition of the
Skorohod integral operator
Z
(u) =
u( \ {t}, t)((dt) (dt)),
(2.2)
X

on any sufficiently integrable measurable process u : X X R. Note that if Dt ut =

0, t X, (u) coincides with the compensated Poisson-Stieltjes integral of u. From
Corollary 1 in [4] we have the duality relation
F Dom(D),

E [hDF, uiL2 (X,) ] = E [F (u)],

u Dom( ).

(2.3)

In addition, for any u Dom( ) we have the commutation relation

Dt (u) = (Dt u) + ut ,

t X.

(2.4)

Moment identities

Using Relations (2.3) and (2.4), the next lemma provides an extension of the Skorohod
isometry to moments of order higher than 2. Here and in other formulas stated in the
sequel we will simply assume that all terms are sufficiently summable and integrable.
Lemma 1. We have
n+1

E [( (u))



Z
n1  
X
n
k
nk+1
(ut )
(dt)
] =
E ( (u))
k
X
k=0
Z

n  
X
n
nk+1
k
k
+
E
(ut )
(( ((I + Dt )u)) ( (u)) )(dt) ,
k
X
k=1

for all n 1.
Proof. This lemma is proved using the identities (2.3) and (2.4) applied to F = ( (u))n .

2

Lemma 1 also shows that the moments of the compensated Poisson stochastic integral
Z
N
+1
\
f (t)((dt) (dt)) of f
Lp (X) satisfy the recurrence identity
X

p=1

" Z
E

n+1 #
f (t)((dt) (dt))

(3.1)

n1   Z
X
n
k=0

(f (t))nk+1 (dt)E

"Z

k #
f (t)((dt) (dt))
.

In particular, in order for (u) to have the same moments as the compensated Poisson
integral of f , it should satisfy the recurrence relation
n+1

E [( (u))

n1   Z
X
n

]=

k=0



(f (t))nk+1 (dt)E ( (u))k ,

(3.2)

n 0, which is Relation (3.1) for the moments of compensated Poisson stochastic integrals, and characterizes their distribution by Carlemans condition when sup kf kLp (Y ) <
p1

. In order to simplify the presentation of moment identities for the Skorohod integral
, it will be convenient to use the symbolic notation
s0 sj

j
Y

usp :=

p=0

where D =

D0 us0 Dj usj ,

(3.3)

0 j ={0,1,...,j}
0
/ 0 , ,j
/ j

Dsj when {0, 1, . . . , j}, j 0, s0 , . . . , sj X.

Let (Y, ) denote another measure space with associated configuration space Y and
Poisson measure with intensity (dy).
Theorem 1. Let N 0 and let R : Lp (Y ) Lp (X) be a random isometry for all
N
+1
\
p = 2, . . . , N + 1. Then for h
Lp (Y ) and n = 0, . . . , N we have
p=2

n+1

E [( (Rh))

]=

n1   Z
X
n
k=0

n X
a  X
n
X
a
+
j b=a
a=0 j=0



(h(y))nk+1 (dy)E ( (Rh))k

C(l0 , . . . , la , a, b, n)

l0 ++la =n+1b
l0 ,...,la 1
la+1 ,...,lb =1

Z
b
Y
q=j+1

!
(h(y))lq (dy) E

Z
t0 tj
X b+1

j
Y

!
(Rh(tp ))lp

#
(dt0 ) (dtj ) ,

p=0

where
C(l0 , . . . , la , a, b, n)


n
ba
= (1)
l0 1 0=r

ba
Y

X
ba+1 <<r0 =b+1

rq 1(ba)+q

q=0 p=rq+1 (ba)+q+1



l1 + + lp q 1
.
l1 + + lp1 q

Proof.
This result is obtained by repeated applications of the integration by parts
formula (2.3) until removal of all terms in .

As a consequence of Theorem 1, if R : Lp (Y ) Lp (X) is a random isometry for all
p = 1, . . . , N + 1, that satisfies the condition
!
Z
j
Y
t0 tj
(Rh(tp ))lp (dt0 ) (dtj ) = 0,
(3.4)
X j+1

p=0

for all l0 + + lj N + 1, l0 1, . . . , lj 1, j = 1, . . . , N , then we have

n+1

E [( (Rh))

]=

n1   Z
X
n

k=0



(h(y))nk+1 (dy)E ( (Rh))k ,

(3.5)

n = 0, . . . , N , i.e. the moments of (Rh) satisfy the recurrence relation (3.1).

Corollary 1. Let R : Lp (Y ) Lp (X) be a random isometry for all p 1, and assume

\
Lp (Y ) satisfies sup khkLp (Y ) < and the cyclic condition
that h
p=1

p1

Dt1 Rh(t2 ) Dtk Rh(t1 ) = 0,

t1 , . . . , tk X,

(3.6)

k -a.e., for all k 2. Then, under , (Rh) has same distribution as the
compensated Poisson integral (h) of h under .
Proof. We first show that (3.6) implies (3.4), and then apply Theorem 1.

Invariance of Poisson measures

Given a measurable random process

: X X Y,
indexed by X, let (), X , denote the image measure of by , i.e.
: X Y
4

(4.1)

maps
(X)

(X)

xi

to

() =

 (xi ) Y .

i=1

i=1
X

In other terms, the random mapping : shifts each configuration point x

according to x 7 (, x). We are interested in finding conditions for : X Y
to map to . This question is well known to have an affirmative answer when the
transformation : X Y is deterministic and maps to .
Corollary 2. Let : X X Y be a measure preserving transformation mapping
to , i.e. (, ) = , X , and satisfying the cyclic condition
Dt1 (, t2 ) Dtk (, t1 ) = 0,

t1 , . . . , tk X,

X ,

(4.2)

for all k 2. Then : X Y maps to , i.e. = is the Poisson measure

with intensity (dy) on Y .
Proof. We apply Corollary 1 to the isometry R : Lp (Y ) Lp (X), p 1, defined
by Rh = h , h Lp (Y ). Then we note that (4.2) implies (3.6) and that we have
(Rh) = (h ) = (h) from (2.2) and the relation Dt Rh(t) = Dt h( (, t)) = 0,
(dt, d)-a.e.

In the above corollary the identity (4.2) is interpreted when Y is a metric space by stating
that for all k 2 and t1 , . . . , tk X the k-tuples
( ( {t1 }, t2 ), ( {t2 }, t3 ), . . . , ( {tk1 }, tk ), ( {tk }, t1 ))
and ( (, t2 ), (, t3 ), . . . , (, tk ), (, t1 )) coincide on at least one component in Y k , for
almost every X . Examples of random transformations : X X Y satisfying
the above hypotheses are considered in [5].

References
[1] A. Dermoune, P. Kree, and L. Wu. Calcul stochastique non adapte par rapport `a la mesure de
Poisson. In Seminaire de Probabilites XXII, volume 1321 of Lecture Notes in Mathematics, pages
477484. Springer Verlag, 1988.
[2] D. Nualart and J. Vives. A duality formula on the Poisson space and some applications. In R. Dalang,
M. Dozzi, and F. Russo, editors, Seminar on Stochastic Analysis, Random Fields and Applications
(Ascona, 1993), volume 36 of Progress in Probability, pages 205213. Birkhauser, Basel, 1995.
[3] G. Di Nunno, B. ksendal, and F. Proske. Malliavin Calculus for Levy Processes with Applications
to Finance. Universitext. Springer-Verlag, Berlin, 2009.
[4] J. Picard. Formules de dualite sur lespace de Poisson. Ann. Inst. H. Poincare Probab. Statist.,
32(4):509548, 1996.
[5] N. Privault. Invariance of Poisson measures under random transformations. Preprint, 2009.
[6] N. Privault. Moment identities for Skorohod integrals on the Wiener space and applications. Electron.
Commun. Probab., 14:116121 (electronic), 2009.

[7] N. Privault. Stochastic Analysis in Discrete and Continuous Settings, volume 1982 of Lecture Notes
in Mathematics. Springer-Verlag, Berlin, 2009.
unel and M. Zakai. Random rotations of the Wiener path. Probab. Theory Relat. Fields,
[8] A.S. Ust
103(3):409429, 1995.
Department of Mathematics, City University of Hong Kong, Tat Chee Avenue, Kowloon
Tong, Hong Kong. nprivaul@cityu.edu.hk