Chapter 4

Annihilation and Creation Operators

In this chapter we present a rst example of a pair of gradient and diver-
gence operators satisfying the duality Assumption 3.1.1, the Clark formula
Assumption 3.2.1 and the stability Assumption 3.2.10 of Section 3.1. This
construction is based on annihilation and creation operators acting on multi-
ple stochastic integrals with respect to a normal martingale. In the following
chapters we will implement several constructions of such operators, respec-
tively when the normal martingale (M
t
)
tR
+
is a Brownian motion or a
compensated Poisson process. Other examples of operators satisfying the
above assumptions will be built in the sequel by addition of a process with
vanishing adapted projection to the gradient D, such as in Section 7.7 on the
Poisson space.
4.1 Duality Relation
The annihilation and creation operators on multiple stochastic integrals pro-
vide a rst concrete example of operators D, satisfying the hypothesis of
Chapter 2. Let the spaces o and | of Section 3.1 be taken equal to
o =
_
n

k=0
I
k
(f
k
) : f
k
L
4
(R
+
)
k
, k = 0, . . . , n, n N
_
, (4.1.1)
and
| =
_
n

i=1
1
[t
i
,t
i1
)
F
i
: F
i
o, 0 = t
0
t
1
< < t
n
, n 1
_
,
which is contained in

| :=
_
n

k=0
I
k
(g
k
(, )) : g
k
L
2
(R
+
)
k
L
2
(R
+
), k = 0, . . . , n, n N
_
,
N. Privault, Stochastic Analysis in Discrete and Continuous Settings,
Lecture Notes in Mathematics 1982, DOI 10.1007/978-3-642-02380-4 4,
c Springer-Verlag Berlin Heidelberg 2009
131
132 4 Annihilation and Creation Operators
where the symmetric tensor product is dened in Section 9.7 of the
Appendix.
In the following denitions the denition of the operators D and are stated
on multiple stochastic integrals (random variables and processes), whose lin-
ear combinations span o and |.
Denition 4.1.1. Let
D : o L
2
( R
+
)
be the linear operator dened by
D
t
I
n
(f
n
) = nI
n1
(f
n
(, t)), dP dt a.e., f
n
L
2
(R
+
)
n
.
Due to its role as a lowering operator on the degree of multiple stochastic
integrals, the operator D is called an annihilation operator in the sequel,
in reference to the use of Fock space expansions (see Denition 2.8.1) in
quantum eld theory.
Denition 4.1.2. Let
:

| L
2
()
be the linear operator dened by
(I
n
(f
n+1
(, ))) = I
n+1
(

f
n+1
), f
n+1
L
2
(R
+
)
n
L
2
(R
+
), (4.1.2)
where

f
n+1
is the symmetrization of f
n+1
in n + 1 variables dened as:

f
n+1
(t
1
, . . . , t
n+1
) =
1
n + 1
n+1

k=1
f
n+1
(t
1
, . . . , t
k1
, t
k+1
, . . . , t
n+1
, t
k
).
In particular we have
f g
n
(t
1
, . . . , t
n+1
) =
1
n + 1
n+1

k=1
f(t
k
)g
n
(t
1
, . . . , t
k1
, t
k+1
, . . . , t
n+1
), (4.1.3)
i.e. f g
n
is the symmetrization of f g
n
in n + 1 variables, cf. Section 9.7.
Similarly to the above, the operator is usually referred to as a creation oper-
ator, due to the fact that it raises the degree of multiple stochastic integrals.
The operator is also called the Skorohod integral.
Note that
(f) = I
1
(f) =
_

0
f(t)dM
t
, f L
2
(R
+
),
and, in particular, from (2.7.4) we have
(uI
n
(f
n
)) = n
_

0
I
n
(f
n
(, s)u

1
[0,s]
n(, ))dM
s
+
_

0
u
s
I
n
(f
n
1
[0,s]
n)dM
s
,
4.1 Duality Relation 133
u L
2
(R
+
), g
n
L
2
(R
+
)
n
, where as a convention denotes the n 1
rst variables and denotes the last integration variable in I
n
. In the next
proposition we show that D and satisfy the duality Assumption 3.1.1.
Proposition 4.1.3. The operators D and satisfy the duality relation
IE[F(u)] = IE[DF, u)
L
2
(R
+
)
], F o, u |. (4.1.4)
Proof. As in Proposition 1.8.2, we consider F = I
n
(f
n
) and u
t
=
I
m
(g
m+1
(, t)), t R
+
, f
n
L
2
(R
+
)
n
, g
m+1
L
2
(R
+
)
m
L
2
(R
+
).
We have
IE[F(u)] = IE[I
m+1
( g
m+1
)I
n
(f
n
)]
= n!1
{n=m+1}
f
n
, g
n
)
L
2
(R
n
+
)
= n!1
{n=m+1}
f
n
, g
n
)
L
2
(R
n
+
)
= n!1
{n1=m}
_

0

_

0
f
n
(s
1
, . . . , s
n1
, t)g
n
(s
1
, . . . , s
n1
, t)ds
1
ds
n1
dt
= n1
{n1=m}
_

0
IE[I
n1
(f
n
(, t))I
n1
(g
n
(, t))]dt
= IE[D

I
n
(f
n
), I
m
(g
m+1
(, )))
L
2
(R
+
)
]
= IE[DF, u)
L
2
(R
+
)
].

In Proposition 4.2.3 below we will show that the Clark formula Assumption
3.2.1 is also satised by D.
Proposition 4.1.4. For any u

| we have
D
t
(u) = u
t
+(D
t
u), t R
+
.
Proof. Letting u
t
= f(t)I
n
(g
n
), t R
+
, f L
2
(R
+
), g
n
L
2
(R
+
)
n
, from
Proposition 4.1.3 we have, by (4.1.3),
D
t
(u) = D
t
(fI
n
(g
n
))
= D
t
I
n+1
(

f g
n
)
= D
t
I
n+1
(f g
n
)
= (n + 1)I
n+1
((f g
n
)(, t))
= f(t)I
n
(g
n
) +nI
n
(f g
n
(, t))
= f(t)I
n
(g
n
) +n(I
n1
(f g
n
(, t)))
= f(t)I
n
(g
n
) +(D
t
I
n
(f g
n
(, t)))
= u
t
+(D
t
u).

134 4 Annihilation and Creation Operators

Remark 4.1.5. By construction, the operator D satises the stability
Assumption 3.2.10 of Chapter 2, thus the conclusion of Proposition 3.2.11
is valid for D, i.e. we have D
s
F = 0, s > t, for any T
t
-measurable F o,
t R
+
.
4.2 Annihilation Operator
From now on we will assume that o is dense in L
2
(), which is equivalent
to saying that (M
t
)
tR
+
has the chaos representation property according to
Denition 2.8.2. As a consequence of Proposition 3.1.2 and Proposition 4.1.3
we have the following.
Proposition 4.2.1. The operators D and are closable in the sense of
Section 9.8 on L
2
() and L
2
( R
+
) respectively.
It also follows from the density of o in L
2
() that | is dense in L
2
(R
+
).
Proposition 4.2.2. The domain Dom(D) = ID([0, )) of D consists in the
space of square-integrable random variables with chaos expansion
F =

n=0
I
n
(f
n
), (4.2.1)
such that the series
n

k=1
kI
k1
(f
k
(, ))
converges in L
2
( R
+
) as n goes to innity.
Given F Dom(D) with the expansion (4.2.1) we have
IE
_
|DF|
2
L
2
(R
+
)
_
=

k=1
kk!|f
k
|
2
L
2
(R
k
+
)
< ,
and
D
t
F = f
1
(t) +

k=1
kI
k1
(f
k
(, t)), dtdP a.e.
In particular, the exponential vector (u), of (2.13.4) belongs to Dom(D) for
all u L
2
(R
+
) and we have
D
s

t
(u) = 1
[0,t]
(s)u(s)
t
(u), s, t [0, T].
4.2 Annihilation Operator 135
The following proposition shows that the Clark formula Assumption 3.2.1 is
satised by D. Its proof parallels the classical argument
f(x) =

n=0

n
x
n
=
0
+

n=1
n
n
_
x
0
y
n1
dy
= f(0) +
_
x
0
f

(y)dy,
described in the introduction for functions of one variable, using the identity
x
n
= n
_
x
0
y
n1
dy.
It shows in particular that the operator D dened in this chapter satises
the Clark formula Assumption 3.2.1 on predictable representation.
Proposition 4.2.3. Every F o can be represented as
F = IE[F] +
_

0
IE[D
t
F [ T
t
]dM
t
. (4.2.2)
Proof. By linearity, in order to prove the statement for F o, it suces to
consider F = I
n
(f
n
). By the denitions of I
n
(f
n
) and D
t
I
n
(f
n
) and using
Lemma 2.7.2 we have, since IE[I
n
(f
n
)] = 0,
I
n
(f
n
) = n
_

0
I
n1
(f
n
(, t)1
[0,t]
n1())dM
t
= n
_

0
IE[I
n1
(f
n
(, t)) [ T
t
]dM
t
=
_

0
IE[D
t
I
n
(f
n
) [ T
t
]dM
t
.

As in the abstract framework of Chapter 3, the Clark formula (4.2.2) extends

to Dom(D) from the closability of D as in Proposition 3.2.3, and to L
2
()
by continuity of F IE[D

F [ T

], cf. Proposition 3.2.6.

Since o dened by (4.1.1) is assumed to be dense in L
2
(), Corollary 3.2.8
and Proposition 4.2.3 show that (M
t
)
tR
+
has the chaos representation prop-
erty as in Denition 2.8.2.
More generally, the following proposition follows from the fact that the dense-
ness of o is equivalent to the chaos representation property.
136 4 Annihilation and Creation Operators
Proposition 4.2.4. If (M
t
)
tR
+
has the chaos representation property then
it has the predictable representation property.
By iteration, the Clark formula (4.2.2) yields the following proposition.
Proposition 4.2.5. For all F
n1
Dom(D
n
) we have
F = IE[F] +

n=1
I
n
(f
n
),
where
f
n
(t
1
, . . . , t
n
) =
1
n!
IE[D
t
1
D
t
n
F],
dt
1
dt
n
dP-a.e., n 1.
Proof. It suces to note that
D
t
1
D
t
n
F = n!f
n
(t
1
, . . . , t
n
) +

k=n+1
k!
(k n)!
I
kn
(f
k
(, t
1
, . . . , t
n
)),
and to use the fact that
IE[I
kn
(f
k
(, t
1
, . . . , t
n
))] = 0, dt
1
dt
n
a.e., k > n 1,
that follows from Proposition 2.7.1 or Lemma 2.7.2.
The above result is analogous to the following expression of Taylors formula:
f(x) = f(0) +

n=1
x
n
n!

n
f
x
n
(0),
with the following correspondence:
calculus on R stochastic analysis
f(x) F
f(0) IE[F]

n
x
n
D
n

n
f
x
n
(0) IE[D
n
F]
The gradient operator D can be extended to the multidimensional case using
the vector-valued multiple stochastic integral (2.14.1).
Denition 4.2.6. Let l 1, . . . , d. We dene the operator
D
(l)
: Dom(D
(l)
) L
2
() L
2
( [0, T])
4.2 Annihilation Operator 137
which maps any F Dom(D
(l)
) with decomposition
F =

n=0
I
n
(f
n
),
to the process (D
(l)
t
F)
t[0,T]
given by
D
(l)
t
F =

n=1
n

h=1
d

i
1
,...,i
n
=1
1
{i
h
=l}
I
n1
(f
i
1
,...,i
n
n
(t
1
, . . . , t
l1
, t, t
l+1
. . . , t
n
)e
i
1
. . . e
i
h1
e
i
h+1
. . . e
i
n
)
=

n=1
nI
n1
(f
l
n
(, t)), dP dt a.e. (4.2.3)
with
f
l
n
= (f
i
1
,...,i
n1
,l
n
e
i
1
. . . e
i
n1
)
1i
1
,...,i
n1
d
.
The domain of D
(l)
is given by
Dom(D
(l)
) =

F =

n=0
d

i
1
,...,i
n
=1
I
n
(f
i
1
,...,i
n
n
e
i
1
. . . e
i
n
) :

n=1
nn!
d

i
1
,...,i
n
=1
|f
i
1
,...,i
n
n
|
2
L
2
([0,T]
n
)
<

.
The Clark formula extends to the multidimensional setting of Section 2.14 as
the next proposition.
Proposition 4.2.7. Let F
d

l=1
Dom(D
(l)
). We have
F = IE[F] +
d

l=1
_
T
0
IE[D
(l)
t
F [ T
t
]dM
(l)
t
. (4.2.4)
In the multidimensional Poisson case we dene the operator D
N
(l)
as in
((4.2.3)) and we have the following Clark formula:
F = IE[F] +
d

l=1
_

0
[
(l)
t
[
1/2
IE[D
N
(l)
t
F [ T
t
](dN
(l)
t

(l)
t
dt),
138 4 Annihilation and Creation Operators
for F

d
l=1
Dom(D
N
(l)
). In the mixed Poisson-Brownian setting of Section
2.14, the operators D
X
(l)
are also dened as in ((4.2.3)) and we have the
Clark formula
F = IE[F] +
d

l=1
_
T
0
IE[D
(l)
t
F [ T
t
]dB
(l)
t
+
p

l=1
_
T
0
IE[D
N
(l)
t
F [ T
t
]dM
(l)
t
,
for F
d+p

l=1
Dom(D
X
(l)
).
4.3 Creation Operator
The domain Dom() of is the space of processes (u
t
)
tR
+
L
2
( R
+
)
with
u
t
=

n=0
I
n
(f
n+1
(, t)),
and such that
IE[[(u)[
2
] =

n=1
(n + 1)!|

f
n
|
2
L
2
(R
n+1
+
)
< .
We will sometimes use the notation
_
b
a
u
s
M
s
:= (1
[a,b]
u), (4.3.1)
to denote the Skorohod integral of u Dom() on the interval [a, b], 0
a b . The creation operator satises the following Ito-Skorohod type
isometry, also called an energy identity for the Skorohod integral.
Proposition 4.3.1. Let u Dom() such that u
t
Dom(D), dt-a.e., and
(D
s
u
t
)
s,tR
+
L
2
( R
2
+
). We have
IE[[(u)[
2
] = IE
_
|u|
2
L
2
(R
+
)
_
+ IE
__

0
_

0
D
s
u
t
D
t
u
s
dsdt
_
, (4.3.2)
Proof. By polarization, orthogonality and density it suces to choose u =
gI
n
(f
n
), f, g L
2
(R
+
), and to note that by the Denition 4.1.2 of we
have
IE[[(u)[
2
] = IE[[(gI
n
(f
n
))[
2
]
= IE[[I
n+1
(f
n
g)[
2
]
4.3 Creation Operator 139
=
1
(n + 1)
2
IE

_
n

i=0
I
n+1
(f
i
g f
(ni)
)
_
2

=
1
(n + 1)
2
_
(n + 1)!(n + 1)|f|
2n
L
2
(R
+
)
|g|
2
L
2
(R
+
)
+n(n + 1)(n + 1)!|f|
2n2
L
2
(R
+
)
f, g)
2
L
2
(R
+
)
)
_
= n!|f|
2n
L
2
(R
+
)
|g|
2
L
2
(R
+
)
+ (n 1)!n
2
|f|
2n2
L
2
(R
+
)
f, g)
2
L
2
(R
+
)
= IE
_
|u|
2
L
2
(R
+
)
_
+g, DI
n
(f
n
))
L
2
(R
+
)
, g, DI
n
(f
n
))
L
2
(R
+
)
)
L
2
()
= IE
_
|u|
2
L
2
(R
+
)
_
+ IE
__

0
_

0
D
s
u
t
D
t
u
s
dsdt
_
.

By polarization, if u and v satisfy the conditions of Proposition 4.3.1 we also

have
(u), (v))
L
2
()
= u, v)
L
2
(R
+
)
+
_

0
_

0
D
s
u
t
, D
t
v
s
)
L
2
()
dsdt.
The proof of (4.3.2) does not depend on the particular type of normal mar-
tingale we are considering, and it can be rewritten as a Weitzenb ock type
identity, cf. [133] and Section 7.6 for details, i.e.:
|(u)|
2
L
2
()
+
1
2
_

0
_

0
|D
s
u
t
D
t
u
s
|
2
L
2
()
dsdt (4.3.3)
= |u|
2
L
2
(R
+
)
+|Du|
2
L
2
(R
2
+
)
.
For Riemannian Brownian motion the study of identities such as (4.3.3) can
be developed via intrinsic dierential operators on Riemannian path space,
cf. [27].
Denition 4.3.2. Let IL
p,1
denote the space of random processes (u
t
)
tR
+
such that u
t
Dom(D), dt-a.e., and
|u|
p
p,1
:= IE
_
|u|
p
L
2
(R
+
)
_
+ IE
__

0
_

0
[D
s
u
t
[
p
dsdt
_
< .
The next result is a direct consequence of Proposition 4.3.1 and Denition
4.3.2 for p = 2.
Proposition 4.3.3. We have IL
2,1
Dom().
As a consequence of Proposition 3.3.1, Proposition 4.1.3 and Proposition
4.2.3, the operator coincides with the Ito integral with respect to (M
t
)
tR
+
on the square-integrable adapted processes, as stated in the next proposition.
140 4 Annihilation and Creation Operators
Proposition 4.3.4. Let (u
t
)
tR
+
L
2
ad
( R
+
) be a square-integrable
adapted process. We have
(u) =
_

0
u
t
dM
t
.
Proof. This result can also be recovered from the denition (4.1.2) of
via multiple stochastic integrals. Since the adaptedness of (u
t
)
tR
+
=
(I
n1
(f
n
(, t)))
tR
+
implies
f
n
(, t) = f
n
(, t)1
[0,t]
n1(), t R
+
,
by Lemma 2.7.2, we have
(I
n
(f
n+1
(, ))) = I
n+1
(

f
n+1
)
= n
_

0
I
n1
(

f
n
(, t)1
[0,t]
n1())dM
t
=
_

0
I
n1
(f
n
(, t)1
[0,t]
n1())dM
t
=
_

0
I
n1
(f
n
(, t))dM
t
, n 1.

Note that when (u

t
)
tR
+
L
2
ad
( R
+
) is a square-integrable adapted pro-
cess, then Relation (4.3.2) becomes the Ito isometry as a consequence of
Proposition 4.3.4, i.e. we have
|(u)|
L
2
()
=
_
_
_
_
_

0
u
t
dM
t
_
_
_
_
L
2
()
(4.3.4)
= |u|
L
2
(R
+
)
, u L
2
ad
( R
+
),
as follows from Remark 4.1.5 since D
t
u
s
= 0, 0 s t, cf. also
Relation (3.3.2) of Proposition 3.3.1.
The following proposition is a Fubini type property for the exchange of
Skorohod and It o stochastic integrals with respect to normal martingales.
Lemma 4.3.5. Let u, v L
4
ad
( R
+
). For all t > 0 we have
_
t
0
u
s
_
t
s
v
r
dM
r
M
s
=
_
t
0
_
r
0
u
s
v
r
M
s
dM
r
, (4.3.5)
where the indenite Skorohod integral is dened in (4.3.1).
4.3 Creation Operator 141
Proof. First, note that
_
r
0
u
s
v
r
M
s
=
_
u

1
{<r}
v
r
_
is T
r
-measurable, r R
+
, hence the stochastic integral in the right hand side
of (4.3.5) exists in the It o sense by Proposition 2.5.4. On the other hand, by
the duality relation (4.1.4) between D and and using the It o-Skorohod
isometry (4.3.4), we have
IE
_
I
n
(f
n
)
_
1
[0,t]
()u

_
t

v
r
dM
r
__
= IE
__
t
0
u
s
_
t
s
v
r
dM
r
D
s
I
n
(f
(n1)
)ds
_
= n
_
t
0
f(s) IE
_
(f()I
n2
(f
(n2)
))(1
[s,t]
u
s
v

)
_
ds
= n
_
t
0
f(s) IE
_
I
n2
(f
(n2)
)u
s
_
t
s
f(r)v
r
dr
_
ds
= nIE
_
I
n2
(f
(n2)
)
_
t
0
f(s)u
s
_
t
s
f(r)v
r
drds
_
= nIE
_
I
n2
(f
(n2)
)
_
t
0
f(r)v
r
_
r
0
f(s)u
s
dsdr
_
= nIE
__
t
0
(u

1
{<r}
v
r
)f(r)drI
n1
(f
(n1)
)
_
= IE
__
t
0
(u

1
{<r}
v
r
)D
r
I
n
(f
n
)dr
_
= IE
_
I
n
(f
n
)
_
t
0
(u

1
{<r}
v
r
)M
r
_
= IE
_
I
n
(f
n
)
_
t
0
(u

1
{<r}
v
r
)dM
r
_
,
for n 1 and f L
2
(R
+
), since the processes u and v are adapted. Hence
by density of o in L
2
() we get

_
1
[0,t]
()u

_
t

v
r
dM
r
_
=
_
t
0

_
u

1
{<r}
v
r
_
dM
r
,
which implies (4.3.5) by (4.3.1).
As a consequence of Proposition 2.11.3 we have the following divergence
formula, whose analog is (1.8.3) in the discrete case. The hypothesis of the
next proposition is satised in particular when L

ad
([0, T] ).
142 4 Annihilation and Creation Operators
Proposition 4.3.6. Suppose that Assumption 2.11.1 holds, i.e.
E
_
_
b
a

2
s
ds

T
a
_
K
2
(b a), P a.s., 0 a b.
Then for all T > 0 and u L

([0, T]) we have

I
1
(u)F = (uF) +u, DF)
L
2
(R
+
)
+(uDF), F o. (4.3.6)
Proof. We prove this result by induction for all F = I
n
(1
[0,t]
nv
n
), n N.
The formula clearly holds for n = 0. From Corollary 2.11.4 we have
I
1
(u1
[0,t]
)I
n
(1
[0,t]
nv
n
)
=
_
t
0
u
s
I
n
(1
[0,s]
nv
n
)dM
s
+n
_
t
0
v
s
I
n1
(1
[0,s]
n1v
(n1)
)
_
s
0
u

dM

dM
s
+n
_
t
0

s
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)dM
s
+n
_
t
0
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)ds.
Applying the induction hypothesis and Relation (4.1.2) we get
I
n1
(1
[0,s]
n1v
(n1)
)
_
s
0
u

dM

= (1
[0,s]
uI
n1
(1
[0,s]
n1v
(n1)
))
+(n 1)
_
s
0
u

dI
n2
(1
[0,s]
n2v
(n2)
)
+(n 1)(1
[0,t]
uvI
n2
(1
[0,s]
n2v
(n2)
))
= I
n
(1
[0,s]
nv
(n1)
u) + (n 1)
_
s
0
u

dI
n2
(1
[0,s]
n2v
(n2)
)
+(n 1)(1
[0,s]
uvI
n2
(1
[0,s]
n1v
(n2)
)),
hence
I
1
(u1
[0,t]
)I
n
(1
[0,t]
nv
n
) =
_
t
0
u
s
I
n
(1
[0,s]
nv
n
)dM
s
+n(n 1)
_
t
0
v
s
_
s
0
u

dI
n2
(1
[0,s]
n2v
(n2)
)dM
s
+n
_
t
0
v
s
I
n
(1
[0,s]
nv
(n1)
u)dM
s
+n(n 1)
_
t
0
v
s
(1
[0,s]
uvI
n2
(1
[0,s]
n1v
(n2)
))dM
s
4.3 Creation Operator 143
+n
_
t
0

s
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)dM
s
+n
_
t
0
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)ds
= I
n+1
(1
[0,t]
n+1v
n
u)
+n(n 1)
_
t
0
v
s
(1
[0,s]
uvI
n2
(1
[0,s]
n1v
(n2)
))dM
s
+n
_
t
0

s
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)dM
s
+n(n 1)
_
t
0
v
s
_
s
0

dI
n2
(1
[0,s]
n2v
(n2)
)dM
s
+n
_
t
0
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)ds
= I
n+1
(1
[0,t]
n+1v
n
u)
+n(n 1)
_
t
0
v
s
(1
[0,s]
uvI
n2
(1
[0,s]
n1v
(n2)
))dM
s
+n
_
t
0

s
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)dM
s
+n
_
t
0
u
s
v
s
dsI
n1
(1
[0,t]
n1v
(n1)
)ds
= (u1
[0,t]
I
n
(1
[0,t]
nv
n
)) +(u1
[0,t]
DI
n
(1
[0,t]
nv
n
))
+u1
[0,t]
, DI
n
(1
[0,t]
nv
n
))
L
2
(R
+
)
,
where in the nal equality we used the relations
I
n1
(1
[0,t]
n1v
(n1)
)
_
t
0
u
s
v
s
ds =
_
t
0

s
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)dM
s
+(n 1)
_
t
0
v
s
_
s
0
u

dI
n2
(1
[0,s]
n2v
(n2)
)dM
s
,
cf. (2.11.7), and
(uv1
[0,t]
I
n1
(1
[0,t]
n1v
(n1)
))
= (n 1)
_
uv1
[0,t]
_
t
0
v
s
I
n2
(1
[0,s]
n2v
(n2)
)dM
s
_
= (n 1)
_

1
[0,t]
()
_
t

v
s
I
n2
(1
[0,s]
n2v
(n2)
)dM
s
_
144 4 Annihilation and Creation Operators
+(n 1)
_

1
[0,t]
()
_

0
v
s
I
n2
(1
[0,s]
n2v
(n2)
)dM
s
_
= (n 1)
_
t
0

1
[0,t]
()
_
t

v
r
I
n2
(1
[0,r]
n2v
(n2)
)dM
r
M
s
+
_

1
[0,t]
()I
n1
(1
[0,]
n1v
(n1)
)
_
= (n 1)
_
t
0
v
s
(1
[0,s]

I
n2
(1
[0,s]
n2v
(n2)
))dM
s
+
_
t
0

s
u
s
v
s
I
n1
(1
[0,s]
n1v
(n1)
)dM
s
,
that follows from Lemma 4.3.5.
4.4 Ornstein-Uhlenbeck Semi-Group
As in the discrete case, a covariance identity can be obtained from the Clark
formula in Section 3.4. In this section we focus on covariance identities ob-
tained from the Ornstein-Uhlenbeck (O.-U.) semi-group (P
t
)
tR
+
dened as
P
t
F =

n=0
e
nt
I
n
(f
n
), (4.4.1)
with F =

n=0
I
n
(f
n
), i.e. P
t
= e
tL
with L = D.
Proposition 4.4.1. Let F, G Dom(D). We have the covariance identity
Cov (F, G) = IE
__

0
_

0
e
s
D
u
FP
s
D
u
Gduds
_
. (4.4.2)
Proof. It suces to prove this identity for F = I
n
(f
n
) and G = I
n
(g
n
) as
Cov (F, G) = IE[I
n
(f
n
)I
n
(g
n
)]
= n!f
n
, g
n
)
L
2
(R
n
+
)
=
1
n
IE
__

0
D
u
FD
u
Gdu
_
= IE
__

0
_

0
D
u
Fe
ns
D
u
Gduds
_
= IE
__

0
e
s
_

0
D
u
FP
s
D
u
Gduds
_
.

4.4 Ornstein-Uhlenbeck Semi-Group 145

Let L
1
denote the inverse of the number operator L = D, cf. Denition
2.8.3, dened on
F L
2
() : IE[F] = 0
as
L
1
F =

n=1
1
n
I
n
(f
n
)
provided F is written as
F =

n=1
I
n
(f
n
).
Note that using the identity
L
1
=
_

0
e
tL
dt =
_

0
P
t
dt,
and the commutation relation DP
t
= e
t
P
t
D, Relation (4.4.2) can also be
obtained from a general semi-group argument:
Cov (F, G) = IE[LL
1
(F IE[F])G]
= IE[DL
1
(F IE[F]), DG)
L
2
(X,)
]
= IE
__

0
DP
t
(F IE[F]), DG)
L
2
(X,)
dt
_
= IE
__

0
e
t
P
t
D(F IE[F]), DG)
L
2
(X,)
dt
_
= IE
__

0
e
t
P
t
DF, DG)
L
2
(X,)
dt
_
.
Relation (4.4.2) implies the covariance inequality
[Cov (F, G)[

IE
_
|DF|
L
2
(R
+
)
_

0
e
s
|P
s
DG|
L
2
(R
+
)
ds
_

|DG|
L

(,L
2
(R
+
))
IE
_
|DF|
L
2
(R
+
)

|DG|
L

(,L
2
(R
+
))
|DF|
L

(,L
2
(R
+
))
,
F, G Dom(D), provided P
s
satises the following continuity property.
Assumption 4.4.2. (Continuity property) For all F Dom(D) we have
|P
t
DF|
L

(,L
2
(R
+
))
|DF|
L

(,L
2
(R
+
))
, t R
+
. (4.4.3)
This property is satised in particular when (
t
)
tR
+
is deterministic, cf.
Section 4.7.
146 4 Annihilation and Creation Operators
4.5 Deterministic Structure Equations
When the process (
t
)
tR
+
is deterministic, Corollary 2.11.4 and Proposition
4.3.6 can be rewritten as a multiplication formula for multiple stochastic
integrals, without requiring any smoothness on (
t
)
tR
+
.
Proposition 4.5.1. Assume that L

(R
+
) is a bounded, deterministic,
function. Then we have
I
1
(u)I
n
(v
n
) = I
n+1
(v
n
u) +nI
n
((uv) v
(n1)
) (4.5.1)
+nu, v)
L
2
(R
+
)
I
n1
(v
(n1)
),
for all u L

(R
+
) L
2
(R
+
), v L
2
(R
+
).
From the above proposition we obtain in particular that for every n 1 there
exists a polynomial Q
n
(x) such that
I
n
(v
n
) = Q
n
(I
1
(v)), n 1, (4.5.2)
see [116] for details when (
s
)
sR
+
is a random process. As seen in Chapters
5 and 6, the Hermite and Charlier polynomials are respectively used to rep-
resent multiple stochastic integrals with respect to Brownian motion and the
compensated Poisson process.
On the other hand, if s
1
< < s
n
and n = n
1
+ +n
d
, we have
1
n
1
[t
0
,t
1
]
1
n
d
[t
d1
,t
d
]
(s
1
, . . . , s
n
)=
n
1
! n
d
!
n!
1
[t
0
,t
1
]
n
1
[t
d1
,t
d
]
n
d
(s
1
, . . . , s
n
),
hence if 0 t
0
< < t
d
,
I
n
(1
n
1
[t
0
,t
1
]
1
n
d
[t
d1
,t
d
]
)
= n!
_

0
_
s
n
0

_
s
2
0
1
n
1
[t
0
,t
1
]
1
n
d
[t
d1
,t
d
]
(s
1
, . . . , s
n
)dM
s
1
dM
s
n
= n
1
! n
d
!
_

0
_
s
n
0

_
s
2
0
1
[t
0
,t
1
]
n
1
[t
d1
,t
d
]
n
d
(s
1
, . . . , s
n
)dM
s
1
dM
s
n
=
d

k=1
_
n
k
!
_

0
_
s
n
k
0

_
s
2
0
1
n
k
[t
k1
,t
k
]
(s
1
, . . . , s
n
k
)dM
s
1
dM
s
n
k
_
=
d

k=1
I
n
k
(1
n
k
[t
k1
,t
k
]
). (4.5.3)
The following is a product rule for the operator D.
4.5 Deterministic Structure Equations 147
Proposition 4.5.2. Assume that L

(R
+
) is a bounded, deterministic
function. We have
D
t
(FG) = FD
t
G+GD
t
F +
t
D
t
FD
t
G, (4.5.4)
t R
+
, F, G o.
Proof. We rst notice that for F = I
1
(u) and G = I
n
(f
n
), this formula is a
consequence of the multiplication formula Proposition 4.5.1 since
D
t
(I
1
(u)I
n
(f
n
))
= D
t
_
I
n+1
(f
n
u) +n
_

0
u
s
I
n1
(f
n
(, s)ds +nI
n
(f
n
(u))
_
= I
n
(f
n
)D
t
I
1
(u) +nI
n
(f
n
(, t) u) +n(n 1)
_

0
u
s
I
n2
(f
n
(, t, s))ds
+n(n 1)I
n
(f
n
(, t) (u)) +
t
D
t
I
1
(u)D
t
I
n
(f
n
)
= I
n
(f
n
)D
t
I
1
(u) +I
1
(u)D
t
I
n
(f
n
) +
t
D
t
I
1
(u)D
t
I
n
(f
n
), t R
+
.
Next, we prove by induction on k N that
D
t
(I
n
(f
n
)(I
1
(u))
k
) = (I
1
(u))
k
D
t
I
n
(f
n
) +I
n
(f
n
)D
t
(I
1
(u))
k
+
t
D
t
(I
1
(u))
k
D
t
I
n
(f
n
),
for all n N. Clearly this formula holds for k = 0. From Proposition 4.5.1
we have
I
n
(f
n
)I
1
(u) H
n1
H
n
H
n+1
, n 1,
hence by the induction hypothesis applied at the rank k we have
D
t
(I
n
(f
n
)(I
1
(u))
k+1
)
= (I
1
(u))
k
D
t
(I
n
(f
n
)I
1
(u)) +I
n
(f
n
)I
1
(u)D
t
(I
1
(u))
k
+
t
D
t
(I
1
(u))
k
D
t
(I
n
(f
n
)I
1
(u))
= (I
1
(u))
k+1
D
t
I
n
(f
n
) +I
n
(f
n
)I
1
(u)D
t
(I
1
(u))
k
+I
n
(f
n
)(I
1
(u))
k
D
t
I
1
(u)
+
t
I
n
(f
n
)D
t
I
1
(u)D
t
(I
1
(u))
k
+
t
I
1
(u)D
t
(I
1
(u))
k
D
t
I
n
(f
n
)
+
t
(I
1
(u))
k
D
t
I
1
(u)D
t
I
n
(f
n
) +
2
t
D
t
I
1
(u)D
t
(I
1
(u))
k
D
t
I
n
(f
n
)
= (I
1
(u))
k+1
D
t
I
n
(f
n
) +I
n
(f
n
)D
t
(I
1
(u))
k+1
+
t
D
t
(I
1
(u))
k+1
D
t
I
n
(f
n
).
Consequently, (4.5.4) holds for any polynomial in single stochastic integrals,
hence from Relation (4.5.2) it holds for any F and G of the formF = I
n
(u
n
),
G = I
n
(v
n
). The extension of F, G o is obtained by an approximation
argument in L
2
() from Proposition (2.11.2).
148 4 Annihilation and Creation Operators
In case
t
= 0, t R
+
, in order for the product relation (4.5.4) of Proposition
4.5.2 to be satised it suces that D
t
be a derivation operator. On the other
hand, if
t
,= 0, t R
+
, Relation (4.5.4) is satised by any nite dierence
operator of the form
F
1

t
(F

t
F).
By induction on r 1 we obtain the following generalization of Relation
(4.5.4).
Corollary 4.5.3. For all F, G o we have
D
t
1
D
t
r
(FG) =
r

p=0
r

q=rp
(4.5.5)

{k
1
<<k
p
}{l
1
<<l
q
}={1,...,r}
D
t
k
1
D
t
k
p
FD
t
l
1
D
t
l
q
G

i{k
1
,...,k
p
}{l
1
,...,l
q
}
(t
i
),
t
1
, . . . t
r
R
+
.
From Proposition 4.5.2, Proposition 4.3.6 can be extended to random u |
as in the next result.
Proposition 4.5.4. Let T R
+
and assume that L

([0, T]) is a locally

bounded deterministic function. Then for all u | and F o we have
(u)F = (uF) +DF, u)
L
2
(R
+
)
+(uDF). (4.5.6)
Proof. The proof of this statement follows by duality from Proposition 4.5.2.
Letting u = vG we have for F, G
1
, G
2
o:
IE[FG
1
(u)] = IE[G
2
v, D(FG
1
))
L
2
(R
+
)
]
= IE[G
2
Fv, DG
1
)
L
2
(R
+
)
] + IE[G
2
G
1
v, DF)
L
2
(R
+
)
]
+IE[G
2
v, DFDG
1
)
L
2
(R
+
)
]
= IE[G
1
(uF)] + IE[G
1
vG
2
, DF)
L
2
(R
+
)
] + IE[G
1
(G
2
vDF)].

If (
t
)
tR
+
is random the probabilistic interpretation of the gradient operator
D is unknown, however we have the following conditional product rule.
Proposition 4.5.5. For F, G o we have
IE[D
t
(FG) [ T
t
] = IE[FD
t
G [ T
t
] + IE[GD
t
F [ T
t
] +
t
IE[D
t
FD
t
G [ T
t
],
F, G o, t R
+
.
4.5 Deterministic Structure Equations 149
Proof. We write (4.5.6) for u | adapted and apply the duality between D
and :
IE[u, D(FG))] = IE[(u)FG]
= IE[G((uF) +u, DF)
L
2
(R
+
)
+(uDF)]
= IE[u, FDG)
L
2
(R
+
)
+u, GDF)
L
2
(R
+
)
+u, DFDG)
L
2
(R
+
)
].

With help of (4.5.4) and Proposition 4.2.5, the following multiplication for-
mula can been proved as a generalization of (4.5.1), cf. [107]. For f
n

L
2
(R
+
)
n
and g
m
L
2
(R
+
)
m
, we dene f
n

l
k
g
m
, 0 l k, to be the
function
(x
l+1
, . . . , x
n
, y
k+1
, . . . , y
m
)
(x
l+1
) (x
k
)
_
R
l
+
f
n
(x
1
, . . . , x
n
)g
m
(x
1
, . . . , x
k
, y
k+1
, . . . , y
m
)dx
1
dx
l
of n + m k l variables. We denote by f
n

l
k
g
m
the symmetrization in
n +mk l variables of f
n

l
k
g
m
, 0 l k.
Proposition 4.5.6. We have the chaos expansion
I
n
(f
n
)I
m
(g
m
) =
2(nm)

s=0
I
n+ms
(h
n,m,s
), (4.5.7)
if and only if the functions
h
n,m,s
=

s2i2(snm)
i!
_
n
i
__
m
i
__
i
s i
_
f
n

si
i
g
m
belong to L
2
(R
+
)
n+ms
, 0 s 2(n m).
Proof. From Corollary 4.5.3 we have
D
t
1
D
t
r
(I
n
(f
n
)I
m
(g
m
)) =
r

p=0
r

q=rp

{k
1
<<k
p
}{l
1
<<l
q
}={1,...,r}
n!
(n p)!
m!
(mq)!
I
np
(f
n
(, t
k
1
, . . . , t
k
p
))I
mq
(g
m
(, t
l
1
, . . . , t
l
q
))

i{k
1
,...,k
p
}{l
1
,...,l
q
}
(t
i
).
150 4 Annihilation and Creation Operators
Dene a function h
n,m,n+mr
L
2
(R
+
)
r
as
h
n,m,n+mr
(t
1
, . . . , t
r
) =
1
r!
IE[D
t
1
D
t
r
(I
n
(f
n
)I
m
(g
m
))]
=
1
r!
rn

p=0
r

q=rp
1
{np=mq}
n!
(n p)!
m!
(m q)!
(n p)!a
n,m,p,r
f
n

np
q+pr
g
m
(t
1
, . . . , t
r
),
=
1
r!

nm+r2p2(nr)
n!m!
(n p)!
a
n,m,p,r
f
n

np
mr+p
g
m
(t
1
, . . . , t
r
),
where a
n,m,p,r
is the number of sequences k
1
< < k
p
and l
1
< < l
q
such
that k
1
, . . . , k
p
l
1
, . . . , l
q
= 1, . . . , r, with exactly mr +p (n p)
terms in common. This number is
a
n,m,p,r
=
r!
(r p)!p!
p!
(mn r + 2p)!(n +r mp)!
.
Hence
h
n,m,n+mr
=

nm+r2p2(nr)
n!m!f
n

np
mr+p
g
m
(r p)!(mn r + 2p)!(n +r mp)!(n p)!
=

n+mr2i2((n+mr)nm)
n!
(n i)!
m!
(mi)!
1
(2i l)!
1
(l i)!
f
n

li
i
g
m
=

l2i2(lnm)
i!
_
n
i
__
m
i
__
i
l i
_
f
n

li
i
g
m
,
with l = n + m r and i = p + m r. The chaos expansion follows from
Proposition 4.2.5, rst for f
n
, g
m
continuous with compact supports. The
general case follows by a density argument.
In the next remark we give a necessary condition for the independence of
multiple stochastic integrals.
Remark 4.5.7. Let f
n
L
2
(R
+
)
n
and g
m
L
2
(R
+
)
m
and assume that
the I
n
(f
n
) and I
m
(g
m
) are independent. Then
_
R
si
+
f
n
(x
1
, . . . , x
n
)g
m
(x
1
, . . . , x
k
, y
i+1
, . . . , y
m
)dx
1
dx
si
= 0, (4.5.8)
(x
si+1
) (x
i
)dx
si+1
dx
n
dy
si+1
dx
m
a.e., 1 2i s 2(n
m).
4.6 Exponential Vectors 151
Proof. If I
n
(f
n
) and I
m
(g
m
) are independent, then I
n
(f
n
)I
m
(g
m
) L
2
()
and
(n +m)! [ f
n
g
m
[
2
L
2
(R
+
)
(n+m)
=[ f
n
g
m
[
2
L
2
(R
+
)
(m+m)
n!m! [ f
n
[
2
L
2
(R
+
)
n
[ g
m
[
2
L
2
(R
+
)
m
= E
_
I
n
(f
n
)
2

E
_
I
m
(g
m
)
2

= E
_
(I
n
(f
n
)I
m
(g
m
))
2
_
=
2(nm)

r=0
(n +mr)! [ h
n,m,r
[
2
L
2
(R
+
)
(n+mr)
= (n +m)! [ f
n
g
m
[
2
L
2
(R
+
)
(n+m)
+
2(nm)

r=1
(n +mr)! [ h
n,m,r
[
2
L
2
(R
+
)
(n+mr)
,
hence h
n,m,r
= 0, r = 1, . . . , 2(n m), which implies (4.5.8).
4.6 Exponential Vectors
We dene a linear transformation T

t
on the space c spanned by the expo-
nential vectors introduced in Denition 2.13.3.
Denition 4.6.1. For all u L
2
(R
+
) let
T

t
(u) = (1 +u
t

t
)(u), u L

(R
+
).
The transformation T

t
is well-dened on c because (u
1
), . . . , (u
n
), are
linearly independent if u
1
, . . . , u
n
are distinct elements of L
2
(R
+
).
Lemma 4.6.2. The transformation T

t
is multiplicative, i.e.
T

t
(FG) = (T

t
F)(T

t
G), F, G c.
Proof. From Lemma 2.13.4 we have
T

t
((u)(v)) = exp(u, v)
L
2
(R
+
)
)T

t
(u +v +uv)
= exp(u, v)
L
2
(R
+
)
)(1 +
t
(u
t
+v
t
+
t
u
t
v
t
))(u +v +uv)
= (1 +
t
u
t
)(1 +
t
v
t
)(u)(v)
= T

t
(u)T

t
(v).

152 4 Annihilation and Creation Operators

The following proposition provides an interpretation of T

t
using the construc-
tion of exponential vectors as solutions of stochastic dierential equations,
cf. Proposition 2.13.1.
Proposition 4.6.3. For all u L
2
(R
+
), T

t

T
(u) coincides dtdP-a.e. with
the limit as T goes to innity of the solution Z
t
T
to the equation
Z
t
s
= 1 +
_
s
0
Z
t

dM
t

, s R
+
, (4.6.1)
where (M
t
s
)
sR
+
is dened as
M
t
s
= M
s
+
t
1
[t,)
(s), s R
+
.
Proof. Clearly by Proposition 2.13.1 we have Z
t
s
=
s
(u), s < t. Next, at
time t we have
Z
t
t
= (1 +
t
u
t
)Z
t
t

= (1 +
t
u
t
)
t
(u)
= (1 +
t
u
t
)
t
(u),
since
t
(u) =
t
(u) a.s. for xed t because M
t
= 0, dt dP-a.e. Finally,
for s > t we have
Z
t
s
= Z
t
t
+
_
s
t
Z
t

dM

= (1 +
t
u
t
)
t
(u) +
_
s
t
Z
t

dM

,
hence
Z
t
s
1 +
t
u
t
=
t
(u) +
_
s
t
Z
t

1 +
t
u
t
u

dM

, s > t,
which implies from (2.13.1):
Z
t
s
1 +
t
u
t
=
s
(u), s > t,
and
Z
t
T
= (1 +
t
u
t
)(u) = T

t
(u),
P-a.s., t R
+
.
In other words, T

t
F, F c, can be interpreted as the evaluation of F on
the trajectories of (M
s
)
sR
+
perturbed by addition of a jump of height
t
at
time t.
In Chapters 5 and 6 we will express the multiple stochastic integrals in terms
of polynomials in the Brownian and Poisson cases. Note that such expressions
using polynomials are not available in other cases, see e.g. [116] in the case
(
t
)
tR
+
is random, in particular for the Azema martingales.
4.6 Exponential Vectors 153
Finally we turn to the probabilistic interpretation of the gradient D. In case
(
t
)
tR
+
is a deterministic function, the probabilistic interpretation of D
t
is
known and D
t
F can be explicitly computed. Dene the operator
D
B
: c L
2
( R
+
, dP dt)
on the space c of exponential vectors as
D
B
F, u)
L
2
(R
+
)
=
d
d
F
_
M() +
_

0
i
s
u
s
ds
_

=0
, F c. (4.6.2)
We have for F = (u) and g L
2
(R
+
):
D
B
F, g)
L
2
(R
+
)
=
d
d
exp
_

_

0
g
s
u
s
i
s
ds
_
(u)

=0
=
_

0
g
s
u
s
i
s
ds (u),
hence D
B
t
(u) = i
t
u
t
(u), t R
+
, where
(u) = exp
__

0
u
s
dM
s

1
2
_

0
[u
s
[
2
i
s
ds
_

sJ
M
(1 +u
s

s
)e
u
s

s
, (4.6.3)
and J
M
denotes the set of jump times of (M
t
)
tR
+
. We have the follow-
ing proposition, which recovers and makes more precise the statement of
Proposition (4.5.2). Let again i
t
= 1
{
t
=0}
and j
t
= 1i
t
= 1
{
t
=0}
, t R
+
.
Proposition 4.6.4. We have
D
t
F = D
B
t
F +
j
t

t
(T

t
F F), t R
+
, F c. (4.6.4)
Proof. When
t
= 0 we have D
B
t
F = i
t
u
t
(u) = i
t
D
t
F, hence
D
t
(u) = i
t
D
t
(u) +j
t
D
t
(u)
= i
t
u
t
(u) +j
t
u
t
(u)
= D
B
t
(u) +
j
t

t
(T

t
(u) (u)), t R
+
.
Concerning the product rule we have from Lemma 2.13.4:
D
t
((u)(v)) = exp
__

0
u
s
v
s
ds
_
D
t
(u +v +uv)
= exp
__

0
u
s
v
s
ds
_
(u
t
+v
t
+
t
u
t
v
t
)(u +v +uv)
154 4 Annihilation and Creation Operators
= (u
t
+v
t
+
t
u
t
v
t
)(u)(v)
= (u)D
t
(v) +(v)D
t
(u) +
t
D
t
(u)D
t
(v),
u, v L

(R
+
), see also Relation (6) of [107].
4.7 Deviation Inequalities
In this section we work under the continuity Assumption 4.4.2, which is
satised when (
t
)
tR
+
is a deterministic function since in this case an
Ornstein-Uhlenbeck process (X
t
)
tR
+
can be associated to the semi-group
(P
s
)
sR
+
. The proof of the next lemma makes forward references to Lemmas
5.3.1 and 6.8.1.
Lemma 4.7.1. The continuity Assumption 4.4.2 is satised if (
t
)
tR
+
is a
deterministic function.
Proof. Let (M
t
)
tR
+
be dened as in (2.10.4) on the product space =
1

2
of independent Brownian motion (B
t
)
tR
+
and Poisson process (N
t
)
tR
+
.
Using the decomposition (2.10.4), i.e.
dM
t
= i
t
dB
t
+
t
(dN
t

t
dt), t R
+
,
any element
G = f(I
1
(u
1
), . . . , I
1
(u
n
))
of o can be constructed as a functional G :
1

2
R. From Lemma 5.3.1
and Lemma 6.8.1 we have
P
t
G() =
_

2
G(T
1
t
(
1
,
1
), T
2
t
(
2
,
2
))p
t
(
1
,
2
, d
1
, d
2
),
for some probability kernel p
t
and mappings
T
1
t
:
1

1

1
, T
2
t
:
1

1

1
.
This implies
|P
t
DF|
L

(,L
2
(R
+
))
|P
t
|DF|
L
2
(R
+
)
|
L

()
|DF|
L

(,L
2
(R
+
))
, t R
+
,
for all F Dom(D).
Proposition 4.7.2. Let F Dom(D) be such that IE[e
T|F|
] < , and e
sF

Dom(D), 0 < s T, for some T > 0. Let h be the function dened by

4.7 Deviation Inequalities 155
h(s) =
_
_
_
_
e
sF
De
sF
DF
_
_
_
_

|DF|
2
L

(,L
2
(R
+
))
, s [0, T].
Then
P(F IE[F] x) exp
_

_
x
0
h
1
(s)ds
_
, 0 < x < h(T),
where h
1
is the inverse of h.
If h is not strictly increasing we may use the left-continuous inverse of h:
h
1
(x) = inf t > 0 : h(t) x, 0 < x < h(T

).
Proof. Assume rst that IE[F] = 0. Since the Ornstein-Uhlenbeck semi-group
(P
t
)
tR
+
satises the continuity Assumption 4.4.2, then using Proposition
4.4.1 we have
IE[Fe
sF
] = Cov (F, e
sF
)
= IE
__

0
e
v
_

0
D
u
e
sF
P
v
D
u
Fdudv
_

_
_
_
_
e
sF
De
sF
DF
_
_
_
_

IE
_
e
sF
_

0
e
v
_

0
D
u
FP
v
D
u
Fdvdu
_

_
_
_
_
e
sF
De
sF
DF
_
_
_
_

IE
_
e
sF
_

0
e
v
|DF|
L
2
(R
+
)
|P
v
DF|
L
2
(R
+
)
dv
_

_
_
_
_
e
sF
De
sF
DF
_
_
_
_

IE
_
e
sF

|DF|
L

(,L
2
(R
+
))
_
_
_
_
_

0
e
v
P
v
|DF|
L
2
(R
+
)
dv
_
_
_
_

_
_
_
_
e
sF
De
sF
DF
_
_
_
_

IE
_
e
sF

|DF|
L

(,L
2
(R
+
))
_

0
e
v
|DF|
L

(,L
2
(R
+
))
dv
IE
_
e
sF

_
_
_
_
e
sF
De
sF
DF
_
_
_
_

|DF|
2
L

(,L
2
(R
+
))
.
h(s) IE
_
e
sF

.
In the general case, letting L(s) = IE[e
s(FIE[F])
], we have
log(IE[e
t(FIE[F])
]) =
_
t
0
L

(s)
L(s)
ds
=
_
t
0
IE[(F IE[F])e
s(FIE[F])
]
IE[e
s(FIE[F])
]
ds
=
_
t
0
h(s)ds,
156 4 Annihilation and Creation Operators
0 t T. We have for all x R
+
:
e
tx
P(F IE[F] x) IE[e
t(FIE[F])
]
e
H(t)
, 0 t T,
where
H(t) =
_
t
0
h(s)ds, 0 t T.
For any 0 < t < T we have
d
dt
(H(t) tx) = h(t) x, hence
min
0<t<T
(H(t) tx) = xh
1
(x) +H(h
1
(x))
= xh
1
(x) +
_
h
1
(x)
0
h(s)ds
= xh
1
(x) +
_
x
0
sdh
1
(s)
=
_
x
0
h
1
(s)ds.

From now on we work with (

t
)
tR
+
a deterministic function, i.e. (M
t
)
tR
+
is
written as in (2.10.4) and from Lemma 4.7.1, the continuity Assumption 4.4.2
is satised.
This covers the Gaussian case for
t
= 0, t R
+
, and also the general Poisson
case when
t
is a non-zero constant.
Proposition 4.7.3. Let K 0 and F Dom(D) be such that
t
D
t
F K,
dtdP-a.e. for some K 0 and |DF|
L

(,L
2
(R
+
))
< . Then we have
P(F IE[F] x) exp
_

|DF|
2
L

(,L
2
(R
+
))
K
2
g
_
xK
|DF|
2
L

(,L
2
(R
+
))
__
exp
_

x
2K
log
_
1 +
xK
|DF|
2
L

(,L
2
(R
+
))
__
, (4.7.1)
x 0, with g(u) = (1 + u) log(1 + u) u, u 0. If K = 0 (decreasing
functionals) we have
P(F IE[F] x) exp
_

x
2
2|DF|
2
L

(,L
2
(R
+
))
_
. (4.7.2)
4.7 Deviation Inequalities 157
Proof. We rst assume that F Dom(D) is a bounded random variable.
Let us assume that IE[F] = 0. From Proposition 4.5.2 we have as in the proof
of Proposition 1.11.1:
0
e
sF
D
u
e
sF
D
u
F
=
1

u
D
u
F
_
e
s
u
D
u
F
1
_

e
sK
1
K
,
since the function x (e
x
1)/x is positive and increasing on R. Hence in
Proposition 4.7.2 we can take
h(s) =

e
sK
1
K

|DF|
2
L

(,L
2
(R
+
))
, s [0, T],
and
min
0<t<T
(H(t) tx) =
_
x
0
h
1
(s)ds

1
K
_
x
0
log
_
1 +tK|DF|
2
L

(,L
2
(R
+
))
_
dt
=
1
K
_
(x +
1
K
|DF|
2
L

(,L
2
(R
+
))
) log
_
1 +xK|DF|
2
L

(,L
2
(R
+
))
_
x
_

x
2K
log
_
1 +
xK
|DF|
2
L

(,L
2
(R
+
))
_
.
If K = 0, the above proof is still valid by replacing all terms by their limits
as K 0. If F Dom(D) is not bounded the conclusion holds for
F
n
= max(n, min(F, n)) Dom(D), n 1,
and (F
n
)
nN
, (DF
n
)
nN
, converge respectively to F and DF in L
2
(), resp.
L
2
( R
+
), with |DF
n
|
2
L

(,L
2
(R
+
))
|DF|
2
L

(,L
2
(R
+
))
.
By the same argument as in Proposition 1.11.3, the bounds (4.7.1) and (4.7.2)
respectively imply
IE[e
|F| log
+
|F|
] <
for some > 0, and
IE[e
F
2
] <
for all < (2|DF|
2
L

(,L
2
(R
+
))
)
1
.
158 4 Annihilation and Creation Operators
Applying Proposition 4.7.3 with the condition DF K for constant
t
=
R
+
, t R
+
, we have the following.
Corollary 4.7.4. Assume that
t
= R
+
, t R
+
, is constant. Let F
Dom(D) be such that DF K for some K 0 and |DF|
L

(,L
2
(R
+
))
< .
Then
P(F IE[F] x) exp
_

|DF|
2
L

(,L
2
(R
+
))

2
K
2
g
_
xK
|DF|
2
L

(,L
2
(R
+
))
__
exp
_

x
2K
log
_
1 +
xK
|DF|
2
L

(,L
2
(R
+
))
__
,
with g(u) = (1 + u) log(1 + u) u, u 0. If = 0 (Wiener case) or K = 0
(decreasing functionals) we have
P(F IE[F] x) exp
_

x
2
2|DF|
2
L

(,L
2
(R
+
))
_
. (4.7.3)
In particular if F is T
T
-measurable, then |DF|
L

(,L
2
(R
+
))
KT and
P(F IE[F] x) exp
_

2
g
_
x
KT
__
exp
_

x
2K
log
_
1 +
x
KT
__
,
which improves (as in [151]) the inequality
P(F IE[F] x) exp
_

x
4K
log
_
1 +
x
2KT
__
, (4.7.4)
which follows from Proposition 6.1 in [6], and relies on modied (not sharp)
logarithmic Sobolev inequalities on Poisson space.
4.8 Derivation of Fock Kernels
In this section we introduce some dierential operators which will used to con-
struct other instances of operators satisfying Assumptions 3.1.1-3.4.3, namely
on the Wiener space in Section 5.8 and on the Poisson space in Section 7.7,
by innitesimal time changes on the paths of the underling process.
4.8 Derivation of Fock Kernels 159
Denition 4.8.1. We dene the linear operator

: o L
2
( R
+
)
on o by

t
I
n
(f
n
) = nI
n
((f

1
[t,)
) f
(n1)
),
t R
+
, f (
1
c
(R
+
), n N, and by polarization of this expression.
The operator

is unbounded, densely dened, and maps functionals of the

n-th chaos H
n
into H
n
, n 1.
For h L
2
(R
+
), let

h
denote the function dened by

h
(t) =
_
t
0
h(s)ds, t R
+
.
Denition 4.8.2. We dene the linear operator

: | L
2
() by

(hI
n
(f
n
)) = nI
n
((f

h
)

f
(n1)
),
f, h (
1
c
(R
+
), and extend it by linearity and polarization.
Next we show that the operators

and

are mutually adjoint.

Proposition 4.8.3. The operators

: o L
2
( R
+
)
and

: | L
2
()
satisfy the duality relation
IE
_

F, u)
L
2
(R
+
)

= IE
_
F

(u)

, F o, u |. (4.8.1)
Proof. By polarization, we need to prove the following. Letting F = I
n
(f
n
),
u = hI
n
(g
n
) and f, g, h (
1
c
(R
+
), we have
IE
_

F, u))
L
2
(R
+
)

= IE
_

I
n
(f
n
), h))
L
2
(R
+
)
I
n
(g
n

= nIE
__

0
I
n
((f

1
[t,)
) f
(n1)
)I
n
(g
n
)h(t)dt
_
= n
2
I
n1
(f
(n1)
), I
n1
(g
(n1)
))
L
2
(R
+
)
(n1)
_

0
h(t)
_

t
f

(s)g(s)dsdt
160 4 Annihilation and Creation Operators
= n
2
(n 1)!f
(n1)
, g
(n1)
)
L
2
(R
+
)
(n1)
_

0
f

(t)g(t)

h
(t)dt
= n
2
(n 1)!f
(n1)
, g
(n1)
)
L
2
(R
+
)
(n1)
_

0
f(t)(

h
g)

(t)dt
= nIE
_
I
n
(f
n
)I
n
((g

h
)

g
(n1)
)
_
= IE
_
I
n
(f
n
)

(hI
n
(g
n
))

= IE
_
F

(u)

,
hence Relation 4.8.1 holds.
Note that the operators

and

are closable from Propositions 3.1.2 and

4.8.3.
4.9 Notes and References
The terminology Skorohod integral is adopted in reference to [136]. The
rst systematic investigation of the relations between the multiple stochas-
tic integrals with respect to normal and the associated annihilation operator
and Skorohod integral appeared in [81]. Proposition 4.2.5 is also known as
the Stroock formula, cf. [138] and Relations (7.4) and (7.5), pages 26-27 of
[66]. We refer to page 216 of [31], and to [68], [140], [141], for other versions
of Proposition 4.5.6 in the Poisson case. In [126] a more general result is
proved, and yields a decomposition the product I
n
(f
n
)I
m
(g
m
) as a sum of
n m integral terms. Those terms are not necessarily linear combinations
of multiple stochastic integrals with respect to (M
t
)
tR
+
, except when the
bracket d[M, M]
t
is a linear deterministic combination of dt and dM
t
, cf.
[116]. Remark 4.5.7 is an extension the necessary condition for independence
proved in the Wiener case in [144]. The necessary and sucient conditions
obtained in [107], [109], [141] are true only when f
n
and g
m
have constant
signs. Necessary and sucient condition for the independence of multiple
stochastic integrals with respect to symmetric -stable random measures with
0 < < 2 have been obtained in [124] as a disjoint support condition on f
n
and g
m
. However, nding a necessary and sucient condition two given sym-
metric functions f
n
and g
m
for the independence of I
n
(f
n
) and I
m
(g
m
) in
the Poisson case is still an open problem. We refer to [101] for the classical
Gaussian deviation inequality (4.7.2) in the case
t
= 0, t R
+
, i.e. on
Wiener space. The material on multidimensional stochastic integrals is taken
from [69]. White noise versions of the annihilation and creation operators,
as well as connections with quantum eld theory can be found in [51]. The
Skorohod isometry Proposition 4.3.1 has also been stated for Brownian mo-
tion on Lie groups and on Riemannian manifolds respectively in [43] and [27].