3

Derivative and Divergence Operators

The Malliavin calculus is a differential calculus on a Gaussian probability

space. In this chapter we introduce the derivative and divergence operators
when the underlying process is a Brownian motion (Bt )t≥0 .

3.1 Finite-Dimensional Case

We consider first the finite-dimensional case. That is, the probability space
(Ω, F , P) is such that Ω = Rn , F = B(Rn ) is the Borel σ-field of Rn , and
P is the standard Gaussian probability with density p(x) = (2π)−n/2 e−|x| /2 .
2

In this framework we consider two differential operators. The first is the

derivative operator, which is simply the gradient of a differentiable func-
tion F : Rn → R:
 ∂F ∂F 
∇F = ,..., .
∂x1 ∂xn
The second differential operator is the divergence operator and is defined
on differentiable vector-valued functions u : Rn → Rn as follows:
n 
 ∂ui 
δ(u) = ui xi − = u, x − div u.
i=1
∂xi

It turns out that δ is the adjoint of the derivative operator with respect to
the Gaussian measure P. This is the content of the next proposition.

Proposition 3.1.1 The operator δ is the adjoint of ∇; that is,

E(u, ∇F) = E(Fδ(u))

if F : Rn → R and u : Rn → Rn are continuously differentiable functions

which, together with their partial derivatives, have at most polynomial
growth.

50
Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 3.2 Malliavin Derivative 51

Proof Integrating by parts, and using ∂p/∂xi = −xi p, we obtain

  n 
∂F
∇F, updx = ui pdx
Rn
i=1 R
n ∂x i

 n    
∂ui
= − F pdx + Fui xi pdx
Rn ∂xi Rn
i=1
= Fδ(u)pdx.
Rn

This completes the proof. 

3.2 Malliavin Derivative

Let B = (Bt )t≥0 be a Brownian motion on a probability space (Ω, F , P) such
that F is the σ-field generated by B. Set H = L2 (R+ ), and for any h ∈ H,
consider the Wiener integral
 ∞
B(h) = h(t)dBt .
0

The Hilbert space H plays a basic role in the definition of the derivative
operator. In fact, the derivative of a random variable F : Ω → R takes
values in H, and (Dt F)t≥0 is a stochastic process in L2 (Ω; H).
We start by defining the derivative in a dense subset of L2 (Ω). More
precisely, consider the set S of smooth and cylindrical random variables of
the form
F = f (B(h1 ), . . . , B(hn )), (3.1)

where f ∈ C ∞
p (R ) and hi ∈ H.
n

Definition 3.2.1 If F ∈ S is a smooth and cylindrical random variable of

the form (3.1), the derivative DF is the H-valued random variable defined
by
 n
∂f
Dt F = (B(h1 ), . . . , B(hn ))hi (t).
i=1
∂x i

For instance, D(B(h)) = h and D(Bt1 ) = 1[0,t1 ] , for any t1 ≥ 0.

This defines a linear and unbounded operator from S ⊂ L2 (Ω) into
L2 (Ω; H). Let us now introduce the divergence operator. Denote by SH
the class of smooth and cylindrical stochastic processes u = (ut )t≥0 of the

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 52 Derivative and Divergence Operators

form

n
ut = F j h j (t), (3.2)
j=1

where F j ∈ S and h j ∈ H.
Definition 3.2.2 We define the divergence of an element u of the form
(3.2) as the random variable given by

n 
n
δ(u) = F j B(h j ) − DF j , h j H .
j=1 j=1

In particular, for any h ∈ H we have δ(h) = B(h).

As in the finite-dimensional case, the divergence is the adjoint of the
derivative operator, as is shown in the next proposition.
Proposition 3.2.3 Let F ∈ S and u ∈ SH . Then
E(Fδ(u)) = E(DF, uH ).
Proof We can assume that F = f (B(h1 ) . . . , B(hn )) and

n
u= g j (B(h1 ) . . . , B(hn ))h j ,
j=1

where h1 , . . . , hn are orthonormal elements in H. In this case, the duality

relationship reduces to the finite-dimensional case proved in Proposition
3.1.1. 
We will make use of the notation Dh F = DF, hH for any h ∈ H and F ∈
S. The following proposition states the basic properties of the derivative
and divergence operators on smooth and cylindrical random variables.
Proposition 3.2.4 Suppose that u, v ∈ SH , F ∈ S, and h ∈ H. Then, if
(ei )i≥1 is a complete orthonormal system in H, we have

∞ 
E(δ(u)δ(v)) = E(u, vH ) + E Dei u, e j H De j v, ei H , (3.3)
i, j=1
Dh (δ(u)) = δ(Dh u) + h, uH , (3.4)
δ(Fu) = Fδ(u) − DF, uH . (3.5)
Property (3.3) can also be written as
 ∞   ∞  ∞ 
E(δ(u)δ(v)) = E ut vt dt + E D s ut Dt v s dsdt .
0 0 0

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 3.3 Sobolev Spaces 53

Proof of Proposition 3.2.4 We first show property (3.4). Consider u =

n
j=1 F j h j , where F j ∈ S and h j ∈ H for j = 1, . . . , n. Then, using
Dh (B(h j )) = h, h j H , we obtain

n 
n 
Dh (δ(u)) = Dh F j B(h j ) − DF j , h j H
j=1 j=1

n 
n
= F j h, h j H + (Dh F j B(h j ) − Dh (DF j ), h j H )
j=1 j=1
= u, hH + δ(Dh u).
To show property (3.3), using the duality formula (Proposition 3.2.3) and
property (3.4), we get
E(δ(u)δ(v)) = E(v, D(δ(u))H )
 ∞ 
=E v, ei H Dei (δ(u))
i=1

∞  
=E v, ei H u, ei H + δ(Dei u)
i=1

∞ 
= E(u, vH ) + E Dei u, e j H De j v, ei H .
i, j=1

Finally, to prove property (3.5) we choose a smooth random variable G ∈ S

and write, using the duality relationship (Proposition 3.2.3),
E(δ(Fu)G) = E(DG, FuH ) = E(u, D(FG) − GDFH )
= E((δ(u)F − u, DFH )G),
which implies the result because S is dense in L2 (Ω). 

3.3 Sobolev Spaces

The next proposition will play a basic role in extending the derivative to
suitable Sobolev spaces of random variables.
Proposition 3.3.1 The operator D is closable from L p (Ω) to L p (Ω; H) for
any p ≥ 1.
Proof Assume that the sequence F N ∈ S satisfies
L p (Ω) L p (Ω;H)
F N −→ 0 and DF N −→ η,

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 54 Derivative and Divergence Operators

as N → ∞. Then η = 0. Indeed, for any u = Nj=1 G j h j ∈ SH such that
G j B(h j ) and DG j are bounded, by the duality formula (Proposition 3.2.3),
we obtain
E(η, uH ) = lim E(DF N , uH )
N→∞
= lim E(F N δ(u)) = 0.
N→∞

This implies that η = 0, since the set of u ∈ SH with the above properties
is dense in L p (Ω; H) for all p ≥ 1. 
We consider the closed extension of the derivative, which we also denote
by D. The domain of this operator is defined by the following Sobolev
spaces. For any p ≥ 1, we denote by D1,p the closure of S with respect to
the seminorm
   ∞  p/2 1/p
F1,p = E(|F| ) + E 
p  2 
(Dt F) dt .
0

In particular, F belongs to D 1,p

if and only if there exists a sequence Fn ∈ S
such that
L p (Ω) L p (Ω;H)
Fn −→ F and DFn −→ DF,
as n → ∞. For p = 2, the space D1,2 is a Hilbert space with scalar product
 ∞ 
F, G1,2 = E(FG) + E Dt FDt Gdt .
0

In the same way we can introduce spaces D1,p (H) by taking the closure of
SH . The corresponding seminorm is denoted by  · 1,p,H .
The Malliavin derivative satisfies the following chain rule.
Proposition 3.3.2 Let ϕ : R → R be a continuous differentiable function
such that |ϕ (x)| ≤ C(1 + |x|α ) for some α ≥ 0. Let F ∈ D1,p for some
p ≥ α + 1. Then, ϕ(F) belongs to D1,q , where q = p/(α + 1), and
D(ϕ(F)) = ϕ (F)DF.
Proof Notice that |ϕ(x)| ≤ C  (1 + |x|α+1 ), for some constant C  , which im-
plies that ϕ(F) ∈ Lq (Ω) and, by Hölder’s inequality, ϕ (F)DF ∈ Lq (Ω; H).
Then, to show the proposition it suffices to approximate F by smooth and
cylindrical random variables, and ϕ by ϕ∗αn , where αn is an approximation
of the identity. 
The chain rule can be extended to the case of Lipschitz functions (see
Exercise 3.3).

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 3.3 Sobolev Spaces 55

We next define the domain of the divergence operator. We identify the

Hilbert space L2 (Ω; H) with L2 (Ω × R+ ).
Definition 3.3.3 The domain of the divergence operator Dom δ in L2 (Ω)
is the set of processes u ∈ L2 (Ω × R+ ) such that there exists δ(u) ∈ L2 (Ω)
satisfying the duality relationship
E(DF, uH ) = E(δ(u)F),
for any F ∈ D1,2 .
Observe that δ is a linear operator such that E(δ(u)) = 0. Moreover, δ is
closed; that is, if the sequence un ∈ SH satisfies
L2 (Ω;H) L2 (Ω)
un −→ u and δ(un ) −→ G,
as n → ∞, then u belongs to Dom δ and δ(u) = G.
Proposition 3.2.4 can be extended to random variables in suitable Sobolev
spaces. Property (3.3) holds for u, v ∈ D1,2 (H) ⊂ Dom δ (see Exercise 3.5)
and, in this case, for any u ∈ D1,2 (H) we can write
 ∞   ∞ ∞ 
E(δ(u) ) ≤ E
2
(ut ) dt + E
2
(D s ut ) dsdt = u21,2,H .
2
0 0 0

Property (3.4) holds if u ∈ D (H) and Dh u ∈ Dom δ (see Exercise 3.6).

1,2

Finally, property (3.5) holds if F ∈ D1,2 , Fu ∈ L2 (Ω; H), u ∈ Dom δ, and

the right-hand side is square integrable (see Exercise 3.7).
We can also introduce iterated derivatives and the corresponding Sobolev
spaces. The kth derivative Dk F of a random variable F ∈ S is the k-
parameter process obtained by iteration:

n
∂k f
Dkt1 ,...tk F = (B(h1 ), . . . , B(hn ))hi1 (t1 ) · · · hik (tk ).
i ,...,i =1
∂xi1 · · · ∂xik
1 k

For any p ≥ 1, the operator Dk is closable from L p (Ω) into L p (Ω; H ⊗k ) (see
Exercise 3.8), and we denote by Dk,p the closure of S with respect to the
seminorm
 k    p/2 1/p
 
j 
j
Fk,p = E(|F| p ) + E (D
 j t1 ,...,t j F)2
dt 1 · · · dt .
j=1 R+

For any k ≥ 1, we set Dk,∞ := ∩ p≥2 Dk,p , D∞,2 := ∩k≥1 Dk,2 , and D∞ :=
∩k≥1 Dk,∞ . Similarly, we can introduce the spaces Dk,p (H).
These spaces satisfy that, for any q ≥ p ≥ 2 and  ≥ k, D,q ⊂ Dk,p . We
also have the following Hölder’s inequality for the  · k,p -seminorms.

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 56 Derivative and Divergence Operators

Proposition 3.3.4 Let p, q, r ≥ 2 such that 1/p + 1/q = 1/r. Let F ∈ Dk,p
and G ∈ Dk,q . Then FG belongs to Dk,r and
FGk,r ≤ c p,q,r Fk,p Gk,q .
Proof The result follows from the Leibnitz rule (see Exercise 3.9) and the
usual Hölder’s inequality. 

3.4 The Divergence as a Stochastic Integral

The Malliavin derivative is a local operator in the following sense. Let
[a, b] ⊂ R+ be fixed. We denote by F[a,b] the σ-field generated by the ran-
dom variables {Bs − Ba , s ∈ [a, b]}.
Lemma 3.4.1 Let F be a random variable in D1,2 ∩ L2 (Ω, F[a,b] , P). Then
Dt F = 0 for almost all (t, ω) ∈ [a, b]c × Ω.
Proof If F belongs to S ∩ L2 (Ω, F[a,b] , P) then this property is clear. The
general case follows by approximation. 
The following result, proved by Gaveau and Trauber (1982), says that
the divergence operator is an extension of Itô’s integral.
Theorem 3.4.2 We have L2 (P) ⊂ Dom δ and, for any u ∈ L2 (P), δ(u)
coincides with Itô’s stochastic integral
 ∞
δ(u) = ut dBt .
0

Proof Consider a simple process u of the form


n−1
ut = φ j 1(t j ,t j+1 ] (t),
j=0

where 0 ≤ t0 < t1 < · · · < tn and the random variables φ j ∈ S are Ft j -

measurable. Then δ(u) coincides with the Itô integral of u because, by (3.5),
n−1 n−1  t j+1
 
n−1
δ(u) = φ j (Bt j+1 − Bt j ) − Dt φ j dt = φ j (Bt j+1 − Bt j ),
j=0 j=0 tj j=0

taking into account that Dt φ j = 0 if t > t j by Lemma 3.4.1. Then the

result follows by approximating any process in L2 (P) by simple processes
(see Proposition 2.1.4), and approximating any φ j ∈ L2 (Ω, Ft j , P) by Ft j -
measurable smooth and cylindrical random variables. 

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 3.5 Isonormal Gaussian Processes 57

If u is not adapted, δ(u) coincides with an anticipating stochastic inte-

gral introduced by Skorohod (1975). Using techniques of Malliavin cal-
culus, Nualart and Pardoux (1988) developed a stochastic calculus for the
Skorohod integral.
If u and v are adapted then, for s < t, Dt v s = 0 and, for s > t, D s ut = 0.
As a consequence, property (3.3) leads to the isometry property of Itô’s
integral for adapted processes u, v ∈ D1,2 (H):
 ∞ 
E(δ(u)δ(v)) = E ut vt dt .
0

If u is an adapted process in D (H) then, from property (3.4), we obtain

1,2

 ∞   ∞
Dt u s dBs = ut + Dt u s dBs , (3.6)
0 t

because Dt u s = 0 if t > s.
The next result shows that we can differentiate Lebesgue integrals of
stochastic processes.
Proposition 3.4.3 For any u ∈ D1,2 (H) and h ∈ H, we have u, hH ∈ D1,2
and
Dt u, hH = Dt u, hH .
Proof Let u ∈ D1,2 (H). Then there exists a sequence un ∈ SH that con-
verges to u in L2 (Ω; H) and is such that the sequence Dun converges to Du
in L2 (Ω; H ⊗ H). Now, let h ∈ H. Then the sequence un , hH converges
to u, hH in L2 (Ω), and the sequence Dun , hH converges to Du, hH in
L2 (Ω; H), which concludes the proof. 
Example 3.4.4 Taking h = 1[0,T ] , we get
 T  T
Dt u s ds = Dt u s ds.
0 0

For example,
 T  T
Dt Bs ds = Dt Bs ds = T − t.
0 0

3.5 Isonormal Gaussian Processes

So far, we have developed the Malliavin calculus with
 ∞respect to Brown-
ian motion. In this case, the Wiener integral B(h) = 0 h(t)dBt gives rise
to a centered Gaussian family indexed by the Hilbert space H = L2 (R+ ).

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 58 Derivative and Divergence Operators

More generally, consider a separable Hilbert space H with scalar prod-

uct ·, ·H . An isonormal Gaussian process is a centered Gaussian family
H1 = {W(h), h ∈ H} satisfying
E(W(h)W(g)) = h, gH ,
for any h, g ∈ H. Observe that H1 is a Gaussian subspace of L2 (Ω). The
notion of isonormal Gaussian process was introduced by Segal (1954).
The Malliavin calculus can be developed in the framework of an isonor-
mal Gaussian process, and all the notions and properties that do not depend
on the fact that H = L2 (R+ ) can be extended to this more general context.
For a complete exposition of the Malliavin calculus with respect to a gen-
eral isonormal Gaussian process we refer to Nualart (2006). We next give
several examples of isonormal Gaussian processes.
Example 3.5.1 Let (T, B, μ) be a measure space, where μ is a σ-finite
measure. Consider a centered Gaussian family of random variables
W = {W(A), A ∈ B, μ(A) < +∞},
with covariance
E(W(A) ∩ W(B)) = μ(A ∩ B).
Then W is called a white noise on (T, B, μ); this is a generalization of the
white noise in D ⊂ Rm (see Definition 1.3.1). The mapping 1A → W(A) can
be extended to a linear isometry from L2 (T ) to the Gaussian space spanned
by W:

ϕ→ ϕ(x)W(dx).
T

Example 3.5.2 Let X = (Xt )t≥0 be a continuous centered Gaussian pro-

cess with covariance function
R(s, t) = E(Xt X s ).
The Gaussian subspace generated by X can be identified with an isonormal
Gaussian process H1 = {X(h), h ∈ H}, where the separable Hilbert space
H is defined as follows. We denote by E the set of step functions on [0, T ].
Let H be the Hilbert space defined as the closure of E with respect to the
scalar product
1[0,t] , 1[0,s] H = R(t, s).
The mapping 1[0,t] → Xt can be extended to an isometry h → X(h) between
H and H1 .

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 3.5 Isonormal Gaussian Processes 59

Example 3.5.3 A particular case of Example 3.5.2 is the fractional Brow-

nian motion, with Hurst parameter H ∈ (0, 1), denoted BH = (BHt )t≥0 .
By definition, BH is a centered Gaussian process with covariance function
given by

RH (t, s) := E(BHt BHs ) = 2−1 (t2H + s2H − |t − s|2H ). (3.7)

This process was introduced by Kolmogorov (1940) and was studied by

Mandelbrot and Van Ness (1968), where a stochastic integral representa-
tion in terms of a two-sided Brownian motion on the whole real line was
established.
One can find a square integrable kernel KH , whose precise expression is
given below, such that
 t∧s
KH (t, u)KH (s, u)du = RH (t, s).
0

This implies the existence

 t of the fractional Brownian through the integral
representation BHt = 0 KH (t, s)dW s , where W = (Wt )t≥0 is a Brownian
motion. Conversely, given BH , we will show that there exists a Brownian
motion W such that this integral representation holds.
Note that, for H = 12 , BH is a standard Brownian motion. However, for
H  12 the increments are non-independent and are positively correlated for
H > 12 and negatively correlated for H < 12 .
The form of the covariance function entails that

E(|BHt − BHs |2 ) = |t − s|2H .

This implies that the process BH has stationary increments. Moreover, by

Kolmogorov’s continuity criterion (Theorem A.4.1), the trajectories of BH
are γ-Hölder continuous on [0, T ] for any γ < H and T > 0. Moreover, the
process is self-similar in the sense that for any a > 0, the processes (BHt )t≥0
and (a−H BHat )t≥0 have the same distribution.
The Gaussian subspace generated by BH can be identified with the iso-
normal Gaussian process defined in Example 3.5.2. However, we do not
know whether the elements of the Hilbert space H introduced in Example
3.5.2 can be considered as real-valued functions. This turn out to be true
for H < 21 but false when H > 12 , as is explained below.
In the case H > 12 , one can show that the kernel KH (t, s) is given by, for
s < t,
 t
KH (t, s) = cH s1/2−H rH−3/2 (r − s)H−1/2 dr,
s

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 60 Derivative and Divergence Operators

where cH is defined as
 H(2H − 1) 1/2
cH = .
β(2 − 2H, H − 12 )
Then, the elements of H may be not functions but distributions of negative
order; see Pipiras and Taqqu (2000). One can show that
L2 ([0, T ]) ⊂ L1/H ([0, T ]) ⊂ |H| ⊂ H,
where |H| is the Banach space of measurable functions ϕ on [0, T ] such
that
 T T
ϕ|H| = H(2H − 1)
2
|r − u|2H−2 |ϕ(r)||ϕ(u)|dudr < ∞.
0 0

In the case where H < the kernel KH (t, s) is given by, for s < t,
1
2
,
 t H−1/2 1   t 
KH (t, s) = cH (t − s)H−1/2
− −H s1/2−H
rH−3/2 (r − s)H−1/2 dr ,
s 2 s

where
 2H 1/2
cH = .
(1 − 2H)β(1 − 2H, H + 12 )
Then, for all α > 1
2
− H, we have that
Cα ([0, T ]) ⊂ H ⊂ L2 ([0, T ]).
Consider the operator KH∗ from E to L2 ([0, T ]) defined as
KH∗ 1[0,t] (s) = KH (t, s)1[0,t] (s).
Then KH∗ is a linear isometry between E and L2 ([0, T ]) that can be extended
to the Hilbert space H.
In the case H > 12 the operator KH∗ can be expressed in terms of fractional
integrals, while in the case H > 12 it can be expressed in terms of fractional
derivatives. In both cases, one can show that, for any a ∈ [0, T ], the indica-
tor function 1[0,a] belongs to the image of KH∗ and thus Im(KH∗ ) = L2 ([0, T ]).
Now, consider the family W = {W(ϕ), ϕ ∈ H} defined as
W(ϕ) = BH ((KH∗ )−1 ϕ).
It is easy to show that W is a family of centered Gaussian random variables
with covariance given by
E(W(ϕ)W(ψ)) = ϕ, ψL2 ([0,T ]) for all ϕ, ψ ∈ L2 ([0, T ]).
In particular, the process Wt = BH ((KH∗ )−1 1[0,t] ) is a Brownian motion.

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 Exercises 61

Moreover, for any ϕ ∈ H, the stochastic integral BH (ϕ) admits the fol-
lowing representation as a Wiener integral:
 T
B (ϕ) =
H
KH∗ ϕ(s)dW s .
0

In particular,
 t
BHt = KH (t, s)dW s .
0

Exercises
3.1 Consider the family P of random variables of the form
p(B(h1 ) . . . , B(hn )), where hi ∈ H and p is a polynomial. Show that P is
dense in Lq (Ω) for all q ≥ 1.
Hint: Assume that q > 1 and let r be the conjugate of q. Show that if Z ∈
Lr (Ω) satisfies E(ZY) = 0 for all Y ∈ P then Z = 0.
3.2 Show that Definition 3.2.1 does not depend on the choice of the representa-
tion for F.
3.3 Let ϕ : R → R be a function such that

|ϕ(x) − ϕ(y)| ≤ K|x − y|,

for all x, y ∈ R. Let F be a random variable in the space D1,2 . Show that ϕ(F)
belongs to D1,2 and that there exists a random variable G bounded by K such
that
D(ϕ(F)) = GDF.

Moreover, show that when the law of the random variable F is absolutely
continuous with respect to the Lebesgue measure, then G = ϕ (F).
Hint: Approximate ϕ by the sequence ϕn = ϕ ∗ αn , where αn is an approxi-
mation of the identity, apply the chain rule in Proposition 3.3.2 to the random
variable F and the function ϕn , and use Corollary 4.2.5 to conclude.
3.4 Let X = (Xt )t∈[0,1] be a continuous centered Gaussian process. Assume that
a.s. X attains its maximum on a unique random point T at [0, 1]. Show that
the random variable M = supt∈[0,1] Xt belongs to the space D1,2 and that
Dt M = 1[0,T ] (t).
Hint: Approximate the supremum of X by the maximum on a finite set, and
use the chain rule for Lipschitz functions (Exercise 3.3), and Corollary 4.2.5.
3.5 Show that, for any u, v ∈ D1,2 (H),
 ∞   ∞  ∞ 
E(δ(u)δ(v)) = E ut vt dt + E D s ut Dt v s dsdt .
0 0 0

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004
 62 Derivative and Divergence Operators

3.6 Suppose that u ∈ D2,2 (H) and let h ∈ H. Show that Dh u belongs to the
domain of the divergence, δ(u) belongs to D1,2 , and
Dh (δ(u)) = δ(Dh u) + h, uH .
3.7 Show that, for any F ∈ D1,2 and u ∈ Dom δ such that Fu ∈ L2 (Ω; H),
δ(Fu) = Fδ(u) − DF, uH ,
provided that the right-hand side is square integrable.
3.8 Show that, for any p ≥ 1, the operator Dk is closable from L p (Ω) into
L p (Ω; H ⊗k ).
3.9 Prove the following Leibnitz rule for an iterated derivative: for any random
variables F, G ∈ S and any k ≥ 2,
k 
k i k−i
Dk (FG) = D FD G.
i=0
i

Downloaded from https://www.cambridge.org/core. University of Warwick, on 18 Sep 2018 at 04:22:01, subject to the Cambridge Core terms of
use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781139856485.004