This action might not be possible to undo. Are you sure you want to continue?
ANALYSIS
Giuseppe Da Prato
June 22, 2009
Contents
1 Gaussian measures in Hilbert spaces 3
1.1 Some concepts of Probability . . . . . . . . . . . . . . . . . . 3
1.1.1 Random variables . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Product measures . . . . . . . . . . . . . . . . . . . . . 5
1.2 Probability measures in Hilbert spaces . . . . . . . . . . . . . 5
1.2.1 Mean and covariance . . . . . . . . . . . . . . . . . . . 5
1.2.2 Finite dimensional projections of measures . . . . . . . 7
1.3 Gaussian probability measures . . . . . . . . . . . . . . . . . . 9
1.3.1 Gaussian probability measures in R . . . . . . . . . . . 9
1.3.2 Gaussian probability measures in R
n
. . . . . . . . . . 10
1.3.3 Gaussian probability measures in H . . . . . . . . . . . 11
1.3.4 Computation of some Gaussian integrals . . . . . . . . 11
1.3.5 The Cameron–Martin space . . . . . . . . . . . . . . . 13
2 Gaussian random variables 17
2.1 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Independent real variables . . . . . . . . . . . . . . . . 18
2.2.2 Independent Gaussian random variables . . . . . . . . 21
2.3 Gaussian random variables deﬁned in a Hilbert space . . . . . 21
2.3.1 Aﬃne changes of variables . . . . . . . . . . . . . . . . 22
2.4 The white noise function . . . . . . . . . . . . . . . . . . . . . 23
2.4.1 Equivalence classes of random variables . . . . . . . . . 23
2.4.2 Deﬁnition of the white noise function . . . . . . . . . . 25
3 Brownian Motion 27
3.1 Stochastic Processes . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Brownian motion . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 Construction of a Brownian motion . . . . . . . . . . . 29
3.2.2 Some properties of a Brownian motion . . . . . . . . . 29
3.3 Wiener integral . . . . . . . . . . . . . . . . . . . . . . . . . . 31
i
ii
3.4 Continuity of Brownian motion . . . . . . . . . . . . . . . . . 35
3.5 The standard Brownian motion . . . . . . . . . . . . . . . . . 36
3.5.1 Some properties of C
0
. . . . . . . . . . . . . . . . . . 37
3.5.2 The Wiener measure and the standard Brownian motion 37
3.6 Quadratic variation of the Brownian motion . . . . . . . . . . 39
3.7 Multidimensional Brownian motions . . . . . . . . . . . . . . . 41
4 Markov property of the Brownian motion 43
4.1 Filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.1.1 F
t
measurable random variables . . . . . . . . . . . . . 44
4.2 Stopping times . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 The Brownian motion W(t + τ) −W(τ) . . . . . . . . . . . . 49
4.4 Transition semigroup . . . . . . . . . . . . . . . . . . . . . . . 50
4.5 Markov property . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5.1 Strong Markov property . . . . . . . . . . . . . . . . . 52
4.6 Some consequences of the strong Markov property . . . . . . . 53
4.7 Application to partial diﬀerential equations . . . . . . . . . . . 56
4.7.1 The Dirichlet problem in the halfline . . . . . . . . . . 57
4.7.2 The Neumann problem . . . . . . . . . . . . . . . . . . 58
4.7.3 The Ventzell problem . . . . . . . . . . . . . . . . . . . 59
5 The Itˆo integral 61
5.1 Deﬁnition of Itˆo’s integral . . . . . . . . . . . . . . . . . . . . 61
5.1.1 Itˆo’s integral for elementary processes . . . . . . . . . . 61
5.1.2 General deﬁnition of Itˆo’s integral . . . . . . . . . . . . 63
5.2 Itˆ o integral for mean square continuous processes . . . . . . . 66
5.3 The Itˆo integral as a stochastic process . . . . . . . . . . . . . 67
5.4 Itˆ o integral with stopping times . . . . . . . . . . . . . . . . . 70
5.4.1 Stopping times . . . . . . . . . . . . . . . . . . . . . . 70
5.4.2 Itˆo’s integral with stopping times . . . . . . . . . . . . 71
5.5 Multidimensional Itˆ o integrals . . . . . . . . . . . . . . . . . . 72
6 The Itˆo formula 75
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.1.1 The Itˆ o formula for unbounded functions . . . . . . . . 82
6.2 Itˆ o’ formula for a vector valued process . . . . . . . . . . . . . 84
7 Stochastic evolution equations 89
7.1 Existence and uniqueness . . . . . . . . . . . . . . . . . . . . . 90
7.1.1 Solution of the stochastic diﬀerential equation in the
space C
B
([s, T]; L
2m
(Ω; R
d
)). . . . . . . . . . . . . . . 94
1
7.1.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.1.3 Diﬀerential stochastic equations with random coeﬃcients 96
7.2 Continuous dependence on data . . . . . . . . . . . . . . . . . 97
7.2.1 Continuous dependence on mean square . . . . . . . . 97
7.3 Almost sure continuity and h¨olderianity of trajectories . . . . 100
7.4 Diﬀerentiability of X(t, s, x) with respect to x . . . . . . . . . 101
7.4.1 Existence of X
x
(t, s, x) . . . . . . . . . . . . . . . . . . 101
7.4.2 Existence of X
xx
(t, s, x) . . . . . . . . . . . . . . . . . 102
7.5 Itˆ o Diﬀerentiability of X(t, s, x) with respect to s. . . . . . . . 105
7.5.1 The deterministic case . . . . . . . . . . . . . . . . . . 105
7.5.2 The stochastic case . . . . . . . . . . . . . . . . . . . . 106
7.5.3 Backward Itˆo’s formula . . . . . . . . . . . . . . . . . . 107
8 Kolmogorov equations 111
8.1 The deterministic case . . . . . . . . . . . . . . . . . . . . . . 111
8.1.1 The autonomous case . . . . . . . . . . . . . . . . . . . 113
8.2 Stochastic case . . . . . . . . . . . . . . . . . . . . . . . . . . 114
8.3 Basic properties of transition operators . . . . . . . . . . . . . 115
8.4 Parabolic equations . . . . . . . . . . . . . . . . . . . . . . . . 116
8.4.1 Autonomous case . . . . . . . . . . . . . . . . . . . . . 117
8.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
A λsystems and πsystems 121
B Conditional expectation 123
B.1 Deﬁnition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
B.2 Basic properties . . . . . . . . . . . . . . . . . . . . . . . . . . 124
C Martingales 127
C.1 Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
C.2 The basic inequality for martingales . . . . . . . . . . . . . . . 128
C.3 Square integrable martingales . . . . . . . . . . . . . . . . . . 129
D Fixed points depending on parameters 133
D.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
D.2 Gˆ ateaux diﬀerentiable mappings . . . . . . . . . . . . . . . . . 134
D.3 The main result . . . . . . . . . . . . . . . . . . . . . . . . . . 135
E Fractional Sobolev spaces and regularity of processes 137
E.1 Fractional Sobolev spaces on [0, 1] . . . . . . . . . . . . . . . . 137
E.2 Processes belonging to W
,2m
(0, T) . . . . . . . . . . . . . . . 138
2
E.3 Multi dimensional Sobolev spaces and regularity of random
ﬁelds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Chapter 1
Gaussian measures in Hilbert
spaces
We shall denote by H a real separable Hilbert space (with inner product
¸, ) and norm [ [), and by L(H) the Banach algebra of all linear bounded
operators T : H → H, endowed with the norm
T = sup
x∈H, x=1
[Tx[.
We recall that T ∈ L(H) is said to be symmetric if ¸Tx, y) = ¸x, Ty) for all
x, y ∈ H, positive if ¸Tx, x) ≥ 0 for all x ∈ H. The set of all symmetric and
positive elements of L(H) will be denoted by L
+
(H).
Finally, we shall denote by C
b
(H) the space of all functions ϕ: H → R
which are continuous and bounded. C
b
(H), endowed with the norm
ϕ
0
: = sup
x∈H
[ϕ(x)[,
is a Banach space.
Next section is devoted to some basic facts from Measure Theory and
Probability needed in what follows.
1.1 Some concepts of Probability
1.1.1 Random variables
Let (Ω, F, P) be a probabilty space and let E be a Polish (complete separable
metric) space; we shall denote by B(E) the σ–algebra generated by all closed
(or equivalently open) subsets of E. The elements of B(E) are called Borel
sets.
3
4 Chapter 1
By an Evalued random variable in (Ω, F) we mean a mapping
X: Ω → E, ω → X(ω),
such that
I ∈ B(E) ⇒ X
−1
(I) ∈ F.
The law (or image measure or pushforward measure) of X is the probability
measure X
#
P on (E, B(E)) deﬁned as
(X
#
P)(I) = P(X
−1
(I)), ∀ I ∈ B(E).
Sometimes we shall use the notation X
#
P = P
X
.
Let us prove the following basic change of variables formula.
Theorem 1.1 Let X be an Evalued random variable in (Ω, F, P). Let
moreover ϕ: E →R be a nonnegative Borel function. Then we have
_
Ω
ϕ(X(ω))P(dω) =
_
E
ϕ(x)(X
#
P)(dx). (1.1)
Proof. Let ﬁrst ϕ = 1l
I
with I ∈ B(E)
(1)
. In this case we have
ϕ(X(ω)) = 1l
X
−1
(I)
(ω), ∀ ω ∈ Ω.
So,
_
Ω
ϕ(X(ω))P(dω) = P(X
−1
(I)) = X
#
P(I) =
_
E
ϕ(x)X
#
P(dx).
Consequently, (1.1) holds for all simple functions ϕ of the form
ϕ =
n
i=1
c
i
1l
I
i
,
with n ∈ N, c
1
, ..., c
n
≥ 0 and I
1
, ..., I
n
∈ B(E). Since any positive Borel
functions is the limit of an increasing sequence of positive simple functions,
the conclusion follows from the monotone convergence theorem.
(1)
1l
I
(ω) is the characteristic function of I; it is equal to 1 if ω ∈ I to 0 if ω / ∈ I.
Gaussian measures 5
1.1.2 Product measures
Let (Ω
i
, F
i
, P
i
), i = 1, ..., n, be probability spaces. Set Ω =
n
i=1
Ω
i
. A mea
surable rectangle of Ω is, by deﬁnition, a set of the form R =
n
i=1
A
i
where
A
i
∈ F
i
, i = 1, 2, ..., n. The σalgebra generated by all measurable rectangles
is called the product σalgebra of F
i
, ..., F
n
; it is denoted by
n
i=1
F
i
.
For any R =
n
i=1
A
i
we deﬁne
P(R) :=
n
i=1
P
i
(A
i
).
One can show that P can be uniquely extended to a probability measure on
(Ω, F) which is called the product probability of P
1
, P
2
, ..., P
n
.
1.2 Probability measures in Hilbert spaces
1.2.1 Mean and covariance
Let µ be a probability measure on (H, B(H)). Assume that µ has ﬁnite ﬁrst
momentum,
_
H
[x[µ(dx) < +∞.
Then the linear functional F : H →R deﬁned as
F(h) =
_
H
¸x, h)µ(dx), ∀ h ∈ H,
is continuous since
[F(h)[ ≤
_
H
[x[µ(dx) [h[, ∀ h ∈ H.
By the Riesz representation theorem there exists m ∈ H such that
¸m, h) =
_
H
¸x, h)µ(dx), ∀ h ∈ H.
m is called the mean of µ. We shall write
m =
_
H
xµ(dx).
Assume now that the second moment of µ is ﬁnite,
_
H
[x[
2
µ(dx) < +∞,
6 Chapter 1
(so that the ﬁrst one is ﬁnite as well). Let us consider the bilinear form
G : H H →R deﬁned as
G(h, k) =
_
H
¸h, x −m)¸k, x −m)µ(dx), ∀ h, k ∈ H.
G is continuous since
[G(h, k)[ ≤
_
H
[x −m[
2
µ(dx) [h[ [k[, ∀ h, k ∈ H.
Therefore there is a unique linear bounded operator Q ∈ L(H) such that
¸Qh, k) =
_
H
¸h, x −m)¸k, x −m)µ(dx), ∀ h, k ∈ H.
Q is called the covariance of µ.
In order to state the next result we need the concept of trace class op
erator. A symmetric and positive operator Q ∈ L(H) is said to be of trace
class if
Tr Q: =
∞
k=1
¸Qe
k
, e
k
) < +∞
for one (and consequently for any) complete orthonormal system (e
k
). One
can show that any trace class operator Q is compact and that Tr Q is the
sum of its eigenvalues repeated according to their multiplicity, see e. g. N.
Dunford and J.T. Schwartz, Linear Operators. Part II, Interscience, 1964.
(2)
Proposition 1.2 The covariance operator Q of µ is symmetric, positive and
of trace class.
Proof. Symmetry and positivity of Q are clear. To prove that Q is of trace
class choose a complete orthonormal system (e
k
) in H. Then we have
¸Qe
k
, e
k
) =
_
H
[¸x −m, e
k
)[
2
µ(dx), k ∈ N.
Therefore, by the monotone convergence theorem and the Parseval identity,
we ﬁnd that
Tr Q =
∞
k=1
_
H
[¸x −m, e
k
)[
2
µ(dx) =
_
H
[x −m[
2
µ(dx) < +∞.
(2)
It is also possible to deﬁne traceclass operators which are not symmetric, but we shall
not need in what follows.
Gaussian measures 7
We shall denote by L
+
1
(H) the set of all positive, symmetric operators in
H of trace class.
We ﬁnally deﬁne the Fourier transform ´ µ of a probability measure µ
setting
´ µ(h) =
_
H
e
ix,h
µ(dx), ∀ h ∈ H. (1.2)
One checks easily that ´ µ : H →C is continuous.
1.2.2 Finite dimensional projections of measures
We are given a probability measure µ ∈ P(H). Let (e
k
) be a complete or
thonormal system in H. For any n ∈ N we consider the projection P
n
: H →
P
n
(H) deﬁned as
P
n
x =
n
k=1
¸x, e
k
)e
k
, x ∈ H. (1.3)
We have lim
n→∞
P
n
x = x for all x ∈ H.
For any n ∈ N we consider the measure µ
n
:= (P
n
)
#
µ deﬁned by
_
H
ϕ(P
n
x)µ(dx) =
_
H
n
ϕ(y)µ
n
(dy),
for all ϕ ∈ C
b
(R).
Thus µ
n
is a probability measure on (P
n
(H), B(P
n
(H)), µ
n
). We shall
also consider µ
n
as a probability measure on (H, B(H), µ), setting
µ
n
(I) = µ
n
(I ∩ P
n
(H)), ∀I ∈ B(H).
We want now to show that µ is determined by the sequence (µ
n
). For this
we ﬁrst need the following result.
Proposition 1.3 Let µ, ν ∈ P(H) be such that
_
H
ϕ(x)µ(dx) =
_
H
ϕ(x)ν(dx), ∀ ϕ ∈ C
b
(H). (1.4)
Then µ = ν.
Proof. Let C ⊂ H be closed and let (ϕ
n
) ⊂ C
b
(H) be such that
(i) lim
n→∞
ϕ
n
(x) = 1l
C
(x) for all x ∈ H.
8 Chapter 1
(ii) ϕ
n

0
≤ 1 for all ∈ N.
A sequence (ϕ
n
) ⊂ C
b
(H) fulﬁlling (i) and (ii) is provided by,
ϕ
n
(x) =
_
_
_
1 if x ∈ C,
1 −n d(x, C) if d(x, C) ≤
1
n
0 if d(x, C) ≥
1
n
.
Now, by the dominate convergence theorem it follows that
lim
n→∞
_
H
ϕ
n
dµ = lim
n→∞
_
H
ϕ
n
dν = µ(C) = ν(C).
Since closed sets generate the Borel σ–algebra of H this implies that µ = ν.
We can now prove the announced result.
Proposition 1.4 Let µ, ν ∈ P(H). If (P
n
)
#
µ = (P
n
)
#
ν for any n ∈ N we
have µ = ν.
Proof. Let ϕ ∈ C
b
(H). Then, using the dominated convergence theorem and
the change of variables formula, we have
_
H
ϕ(x)µ(dx) = lim
n→∞
_
H
ϕ(P
n
x)µ(dx) = lim
n→∞
_
P
n
(H)
ϕ(ξ)((P
n
)
#
µ)(dξ)
and
_
H
ϕ(x)ν(dx) = lim
n→∞
_
H
ϕ(P
n
x)ν(dx) = lim
n→∞
_
P
n
(H)
ϕ(ξ)((P
n
)
#
ν)(dξ).
Since (P
n
)
#
µ = (P
n
)
#
ν by assumption, we conclude that
_
H
ϕ(x)µ(dx) =
_
H
ϕ(x)ν(dx)
for all ϕ ∈ C
b
(H). Therefore, in view of Proposition 1.3 we have µ = ν.
As an application of Proposition 1.4 we prove that the Fourier transform
of µ determines µ.
Proposition 1.5 Let µ, ν ∈ P(H) be such that ´ µ(h) = ´ ν(h) for all h ∈ H.
Then µ = ν.
Gaussian measures 9
Proof. We assume as granted the result when H is ﬁnitedimensional
(3)
. In
the general case we have by (1.1) for any h ∈ H and n ∈ N,
´ µ(P
n
h) =
_
H
e
ix,P
n
h
µ(dx) =
_
P
n
(H)
e
iP
n
ξ,P
n
h
(P
n
)
#
µ(dξ) =
(P
n
)
#
µ(P
n
h)
and
´ ν(P
n
h) =
_
H
e
ix,P
n
h
ν(dx) =
_
P
n
(H)
e
iP
n
ξ,P
n
h
(P
n
)
#
ν(dξ) =
(P
n
)
#
ν(P
n
h).
Therefore measures (P
n
)
#
µ and (P
n
)
#
ν have the same Fourier tranforms and
so they coincide. The conclusion follows from Proposition 1.4.
1.3 Gaussian probability measures
We ﬁrst recall the deﬁnition of Gaussian measure on (R, B(R)), then we go
to the general case.
1.3.1 Gaussian probability measures in R
For any pair of real numbers (m, q) with m ∈ R and q ≥ 0 we deﬁne a
probability measure N
m,q
on (R, B(R)) as follows. If q = 0 we set
N
m,0
= δ
m
,
where δ
m
is the Dirac measure at m, deﬁned for all B ∈ B(R) by
δ
m
(B) =
_
_
_
1 if m ∈ B,
0 if m / ∈ B.
If q > 0 we set
N
m,q
(B) =
1
√
2πq
_
B
e
−
(x−m)
2
2q
dx, for all B ∈ B(R).
N
m,q
is a probability measure since
N
m,q
(R) =
1
√
2πq
_
+∞
−∞
e
−
(x−m)
2
2q
dx =
1
√
2π
_
+∞
−∞
e
−
x
2
2
dx = 1.
(3)
See e.g. M. M´etivier, Notions fondamentales de la th´eorie des probabilit´ees, Dunod
Universit´e, 1968.
10 Chapter 1
If q > 0, N
m,q
is absolutely continuous with respect to the Lebesgue measure
1
(dx) = dx in (R, B(R) and
N
m,q
(dx) =
1
√
2πq
e
−
(x−m)
2
2q
dx.
When m = 0 we shall write for short N
q
instead N
0,q
.
It is easy to see that m is the mean and q the covariance of N
m,q
. Moreover,
its Fourier transform is given by
N
m,q
(h) :=
_
R
e
ihx
N
m,q
(dx) = e
imh−
1
2
qh
2
, h ∈ R. (1.5)
1.3.2 Gaussian probability measures in R
n
We are going to deﬁne a Gaussian measure N
m,Q
for any m = (m
1
, ..., m
n
) ∈
R
n
and any Q ∈ L
+
(R
n
).
Let Q ∈ L
+
(R
n
) and let (e
1
, ..., e
n
) be an orthonormal basis on R
n
such
that Qe
k
= λ
k
e
k
, k = 1, ..., n, for some λ
k
≥ 0. Then we deﬁne a probability
measure N
a,Q
on (R
n
, B(R
n
)) by setting
N
m,Q
=
n
k=1
N
m
k
,λ
k
.
When m = 0 we shall write N
Q
instead of N
m,Q
for short.
The proof of the following proposition is easy; it is left to the reader.
Proposition 1.6 Let m ∈ R
n
, Q ∈ L
+
(R
n
) and µ = N
m,Q
. Then we have
_
R
n
xµ(dx) = m,
_
R
n
¸y, x −a)¸z, x −a)µ(dx) = ¸Qy, z), y, z ∈ R
n
.
Moreover the Fourier tranform of N
a,Q
is given by
¯
N
a,Q
(h) :=
_
R
n
e
ih,x
µ(dx) = e
ia,h−
1
2
Qh,h
, h ∈ R
n
.
Finally, if the determinant of Q is positive, N
a,Q
is absolutely continuous
with respect to the Lebesgue measure in R
n
and we have
N
a,Q
(dx) =
1
_
(2π)
d
det Q
e
−
1
2
Q
−1
(x−a),x−a
dx.
Therefore m is the mean and Q the covariance operator of N
a,Q
.
Gaussian measures 11
1.3.3 Gaussian probability measures in H
Let m ∈ H and Q ∈ L
+
1
(H). We denote by N
m,Q
the probability measure on
(H, B(H)) of mean m, covariance Q and Fourier transform given by
N
m,Q
(h) = e
im,h−
1
2
Qh,h
, h ∈ H. (1.6)
One can show that such a measure does exist
(4)
; it is unique thank’s to
Proposition 1.5.
1.3.4 Computation of some Gaussian integrals
To compute some integrals with respect to a Gaussian measure µ = N
m,Q
in
an inﬁnite dimensional Hilbert space H it is useful to reduce the computation
to integrals on a sequence (H
n
) of ﬁnite dimensional vector spaces convergent
to H and then to let n → ∞.
More precisely, given µ = N
m,Q
∈ P(H), we shall proceed as follows.
Since Q is compact there exists an orthonormal complete system (e
k
) in H
and a sequence of nonnegative numbers (λ
k
) such that
Qe
k
= λ
k
e
k
, ∀ k ∈ N.
For any n ∈ N we set m
n
:= ¸m, e
n
),
P
n
x =
n
k=1
¸x, e
k
)e
k
, ∀ x ∈ H
and identify P
n
(H) with R
n
through the isomorphism,
P
n
(H) →R
n
, x =
n
k=1
¸x, e
k
)e
k
→ (¸x, e
1
), ..., ¸x, e
n
)).
Exercise 1.7 Prove that
µ
n
= (P
n
)
#
µ =
n
i=1
N
m
k
,λ
k
.
Hint. Show that the Fourier transform of µ
n
is given by
´ µ
n
(h) = e
i
P
n
k=1
m
k
h
k
e
−
1
2
P
n
k=1
λ
k
h
2
k
.
(4)
see e.g. G. Da Prato, An introduction to inﬁnitedimensional analysis. Springer
Verlag, Berlin, 2006.
12 Chapter 1
We shall assume (which is always true after a rearrangement) that λ
1
≥
λ
2
≥ λ
n
≥ .
To formulate the next result notice that for any ε <
1
λ
1
, the linear operator
1 − εQ is invertible and (1 − εQ)
−1
is bounded. We have in fact, as easily
checked,
(1 −εQ)
−1
x =
∞
k=1
1
1 −ελ
k
¸x, e
k
)e
k
, x ∈ H.
In this case we can deﬁne the determinant of (1 −εQ) by setting
det(1 −εQ): = lim
n→∞
n
k=1
(1 −ελ
k
) :=
∞
k=1
(1 −ελ
k
).
Exercise 1.8 Prove that
∞
k=1
(1 −ελ
k
) > 0.
Hint. Write
log
_
∞
k=1
(1 −ελ
k
)
_
=
∞
k=1
log(1 −ελ
k
)
and show that the series is convergent since
∞
k=1
λ
k
< +∞.
Proposition 1.9 Let ε ∈ R. Then we have
_
H
e
ε
2
x
2
µ(dx) =
_
_
_
[det(1 −εQ)]
−1/2
e
ε
2
(1−εQ)
−1
m,m
, if ε <
1
λ
1
,
+∞, otherwise.
(1.7)
Proof. For any n ∈ N we have, taking into account Exercise 1.7
_
H
e
ε
2
P
n
x
2
µ(dx) =
_
P
n
(H)
e
ε
2
P
n
ξ
2
µ
n
(dξ) =
n
k=1
_
R
e
ε
2
ξ
2
k
N
m
k
,λ
k
(dξ
k
).
Since [P
n
x[
2
↑ [x[
2
as n → ∞ and, by an elementary computation,
_
R
e
ε
2
x
2
k
N
m
k
,λ
k
(dx
k
) =
1
√
1 −ελ
k
e
−
ε
2
m
2
k
1−ελ
k
,
the conclusion follows from the monotone convergence theorem.
Gaussian measures 13
Exercise 1.10 Prove that for all m ∈ N
J
m
:=
_
H
[x[
2m
µ(dx) < ∞
and compute J
m
.
Hint. Notice that J
m
= 2
m
F
(m)
(0), where
F(ε) =
_
H
e
ε
2
x
2
µ(dx), ε > 0.
Proposition 1.11 We have
_
H
e
h,x
µ(dx) = e
a,h
e
1
2
Qh,h
, h ∈ H. (1.8)
Proof. For any ε > 0 we have
e
h,x
≤ e
x h
≤ e
εx
2
e
1
ε
h
2
.
Choosing ε <
1
λ
1
, we have, by the dominated convergence theorem, that
_
H
e
h,x
µ(dx) = lim
n→∞
_
H
e
h,P
n
x
µ(dx) = lim
n→∞
_
P
n
(H)
e
h,P
n
ξ
µ
n
(dx)
= lim
n→∞
e
P
n
m,h
e
1
2
P
n
Qh,h
= e
m,h
e
1
2
Qh,h
.
1.3.5 The Cameron–Martin space
We are given a Gaussian measure µ = N
Q
, where Q ∈ L
+
1
(H). We say that
µ is non degenerate if Ker Q := ¦x ∈ H : Qx = 0¦ = ¦0¦. Thus, if H is
ﬁnitedimensional µ is non degenerate if and only if det Q > 0.
Assume now that H is inﬁnitedimensional and that µ is non degenerate.
We denote by (e
k
) a complete orthonormal system in H such that Qe
k
=
λ
k
e
k
, k ∈ N, where (λ
k
) are the eigenvalues of Q and we set x
k
= ¸x, e
k
), k ∈
N.
We notice that the inverse Q
−1
of Q (which is well deﬁned since Ker
Q = ¦0¦) is not continuous because,
Q
−1
e
k
=
1
λ
k
e
k
, k ∈ N
and λ
k
→ 0 as k → ∞. Consequently, recalling the closed graph theorem,
we see that the range Q(H) does not coincide with H. However, it is dense
in H as the following lemma shows.
14 Chapter 1
Lemma 1.12 Q(H) is a dense subspace of H.
Proof. In fact if x
0
is an element of H orthogonal to Q(H), we have
¸Qx, x
0
) = ¸x, Qx
0
) = 0, ∀ x ∈ H,
which yields Qx
0
= 0, and so x
0
= 0 because Ker(Q) = ¦0¦.
It is useful to introduce the operator Q
1/2
deﬁned as
Q
1/2
x =
∞
k=1
_
λ
k
¸x, e
k
)e
k
, x ∈ H.
Its range Q
1/2
(H) is called the Cameron–Martin space of the measure µ.
Arguing as before we see that Q
1/2
(H) is a subspace of H diﬀerent of H and
dense in H. Moreover it is clear that x ∈ Q
1/2
(H) if and only if,
∞
k=1
λ
−1
k
x
2
k
< +∞.
It is important to notice that the measure of the Cameron–Martin space
is zero.
Proposition 1.13 We have µ(Q
1/2
(H)) = 0.
Proof. For any n, k ∈ N set
U
n
=
_
y ∈ H :
∞
h=1
λ
−1
h
y
2
h
< n
2
_
= ¦y ∈ Q
1/2
(H) : [Q
−1/2
y[ < n¦,
and
U
n,k
=
_
y ∈ H :
2k
h=1
λ
−1
h
y
2
h
< n
2
_
.
Clearly U
n
↑ Q
1/2
(H) as n → ∞, and for any n ∈ N, U
n,k
↓ U
n
as k → ∞.
So, it is enough to show that
µ(U
n
) = lim
k→∞
µ(U
n,k
) = 0. (1.9)
We have in fact
µ(U
n,k
) =
_
¦
y∈H:
P
2k
h=1
λ
−1
h
y
2
h
<n
2
¦
2k
h=1
N
λ
k
(dy
k
),
Gaussian measures 15
which, setting z
h
= λ
−1/2
h
y
h
is equivalent to
µ(U
n,k
) =
_
¦
z∈R
2k
:z<n
¦
N
I
2k
(dz),
where I
2k
is the identity in R
2k
. Let us compute µ(U
n,k
). We have
µ(U
n,k
) =
µ(U
n,k
)
µ(H)
=
_
n
0
e
−
r
2
2
r
2k−1
dr
_
+∞
0
e
−
r
2
2
r
2k−1
dr
=
_
n
2
/2
0
e
−ρ
ρ
k−1
dρ
_
+∞
0
e
−ρ
ρ
k−1
dρ
.
Therefore
µ(U
n,k
) =
1
(k −1)!
_
n
2
/2
0
e
−ρ
ρ
k−1
dρ ≤
1
(k −1)!
_
n
2
/2
0
ρ
k−1
dρ =
1
k!
_
n
2
2
_
k
,
and (1.9) follows.
16 Chapter 1
Chapter 2
Gaussian random variables
2.1 Notations
Let (Ω, F, P) be a probability space, H a separable Hilbert space, X: Ω → H
a random variable such that
_
Ω
[X(ω)[
2
P(dω) < ∞.
We denote by X
#
P the law of X, by m(X) the mean of X
#
P and by Q(X)
the covariance of X
#
P.
By the change of variables formula it follows that the Fourier transform
of X
#
P is given by
X
#
P(h) =
_
Ω
e
iX(ω),h
P(dω), ∀ h ∈ H
and that
¸m(X), h) =
_
Ω
¸X(ω), h)P(dω), ∀ h ∈ H,
and
¸Q(X)h, k) =
_
Ω
¸X(ω) −m(X), h) ¸X(ω) −m(X), k)P(dω), ∀ h, k ∈ H.
Deﬁnition 2.1 We say that X
#
P is a Gaussian random variable if X
#
P is
a Gaussian measure, that is if
X
#
P(h) = e
im(X),h
e
−
1
2
Q(X)h,h
, ∀ h ∈ H.
In this case we call m(X) the mean and Q(X) the covariance of X.
17
18 Chapter 2
Example 2.2 Let n ∈ N, X
1
, ..., X
n
be real random variables on (Ω, F, P).
Then X = (X
1
, ..., X
n
) is a R
n
valued random variable. So, m(X) is a vector
of R
n
denoted by (m(X)
1
, ..., m(X)
n
) and Q(X) is a n n matrix denoted
Q(X)
i,j
, i, j = 1, ..., n.
More precisely, let (e
1
, ..., e
n
) be the canonical basis in R
n
. Then for any
k = 1, ..., n we have
m(X)
k
= ¸m(X), e
k
) =
_
Ω
X
k
(ω)P(dω) = m(X
k
)
and for any j, k = 1, ..., n we have
Q(X)
j,k
= ¸Q(X)e
j
, e
k
) =
_
Ω
(X
j
(ω) −m
j
(X
j
))(X
k
(ω) −m
k
(X
k
))P(dω).
In particular, if j = k we ﬁnd
Q(X)
k,k
= Q(X
k
), k = 1, ..., n.
Example 2.3 Assume that X = (X
1
, ..., X
n
) is a ndimensional Gaussian
random variable. Then X
1
, ..., X
n
are real Gaussian random variables. In
fact if k = 1, ..., n and a ∈ R we have
_
Ω
e
iaX
k
(ω)
P(dω) =
_
Ω
e
iae
k
,X(ω)
P(dω)
= e
iae
k
,m(X)
e
−
1
2
a
2
Q(X)e
k
,e
k
= e
iam(X
k
)
e
−
1
2
a
2
Q(X
k
)
.
Notice that, if conversely X
1
, ..., X
n
are real Gaussian random variables, then
X = (X
1
, ..., X
n
) is not necessarily Gaussian.
2.2 Independence
In this section we introduce the basic concept of independence.
2.2.1 Independent real variables
Deﬁnition 2.4 Let n ∈ N and let X
1
, ..., X
n
be real random variables in
(Ω, F, P). Consider the R
n
valued random variable
X(ω) = (X
1
(ω), ..., X
n
(ω)), ω ∈ Ω.
random variables 19
We say that X
1
, ..., X
n
are independent if
X
#
P =
n
j=1
(X
j
)
#
P.
Let (X
i
) be a sequence of real random variables. They are called independent
if X
i
1
, . . . , X
i
n
are independent for any choice of n and of positive integers
i
1
< i
2
< < i
n
.
A necessary and suﬃcient condition for the independence is provided by
the following proposition.
Proposition 2.5 Let X
1
, ..., X
n
, n ∈ N, be real independent random vari
ables in (Ω, F, P). Let moreover ϕ
1
, ..., ϕ
n
be Borel positive functions. Then
we have
_
Ω
ϕ
1
(X
1
(ω)) ϕ
n
(X
n
(ω))P(dω)
=
_
Ω
ϕ
1
(X
1
(ω))P(dω)
_
Ω
ϕ
n
(X
n
(ω))P(dω).
(2.1)
Conversely, if (2.1) holds for any choice of positive Borel functions ϕ
1
, ..., ϕ
n
,
then X
1
, ..., X
n
are independent.
Proof. Set X = (X
1
, ..., X
n
) and let ψ: R
n
→R be deﬁned as
ψ(ξ
1
, ..., ξ
n
) = ϕ
1
(ξ
1
) ϕ
k
(ξ
n
), (ξ
1
, ..., ξ
n
) ∈ R
n
.
Then by the change of variable formula we have, taking into account the
independence of X
1
, ..., X
n
,
_
Ω
ϕ
1
(X
1
(ω)) ϕ
n
(X
n
(ω))P(dω) =
_
Ω
ψ(X(ω))P(dω)
=
_
R
n
ψ(ξ)(X
#
P)(dξ) =
_
R
ϕ
1
(ξ
1
)((X
1
)
#
P)(dξ
1
)
_
R
ϕ
k
(ξ
n
)((X
n
)
#
P)(dξ
n
)
=
_
Ω
ϕ
1
(X
1
(ω))P(dω)
_
Ω
ϕ
n
(X
n
(ω))P(dω).
Assume conversely that (2.1) holds for any choice of functions ϕ
1
, ..., ϕ
n
positive Borel. To prove independence of X
1
, ..., X
n
it is enough to show that
(X
#
P)(I
1
I
n
) = ((X
1
)
#
P)(I
1
) ((X
n
)
#
P)(I
n
), ∀ I
1
, ..., I
n
∈ B(R).
But this follows immediately setting in (2.1)
ϕ
i
= 1l
I
i
, i = 1, ..., n.
20 Chapter 2
Exercise 2.6 Let X
1
, ..., X
n
be real independent random variables in (Ω, F, P).
Show that
_
Ω
X
1
X
n
dP =
_
Ω
X
1
dP
_
Ω
X
n
dP
and
V (X
1
+ + X
n
) = V (X
1
) + + V (X
n
).
The following useful result is left to the reader as an exercise.
Proposition 2.7 Let X
1
, ..., X
n
be real random variables in (Ω, F, P) and
let X = (X
1
, ..., X
n
). Then X
1
, ..., X
n
are independent if and only if
X
#
P(h) =
n
k=1
(X
k
)
#
P(h
k
), ∀ h = (h
1
, ..., h
n
) ∈ R
n
.
Deﬁnition 2.8 Let (Ω, F, P) be a probability space and A
1
, ..., A
n
∈ F.
We say that the sets A
1
, ..., A
n
are independent if the random variables
1l
A
1
, ..., 1l
A
n
are so.
Exercise 2.9 Show that sets A
1
, ..., A
n
are independent if and only if
P(A
j
1
∩ ∩ A
j
k
) = P(A
j
1
) P(A
j
k
),
for all k = 1, ..., n and k diﬀerent positive integer j
1
, ..., j
k
less or equal to n.
Proposition 2.10 Let X
1
, ..., X
n
be real independent random variables in
(Ω, F, P) and let X = (X
1
, ..., X
n
). Then the covariance matrix Q(X) is
diagonal.
Proof. We have in fact (by Exercise 2.6) for i, j = 1, ..., n
Q(X)
i,j
=
_
Ω
(X
i
(ω) −m
i
(X))(X
j
(ω) −m
j
(X))P(dω)
=
_
Ω
(X
i
(ω) −m
i
(X))P(dω)
_
Ω
(X
j
(ω) −m
j
(X))P(dω) = 0.
The converse of Proposition 2.10 does not hold in general.
random variables 21
2.2.2 Independent Gaussian random variables
Let X
1
, ..., X
n
be real random variables in (Ω, F, P) and let X = (X
1
, ..., X
n
).
Proposition 2.11 Assume that X
1
, ..., X
n
are independent Gaussian ran
dom variables. Then X = (X
1
, ..., X
n
) is Gaussian.
Proof. In fact, let h = (h
1
, ..., h
n
) ∈ R
n
. Then, taking into account the
independence of (X
1
, ..., X
n
),
X
#
P(h) =
_
Ω
e
i(X
1
(ω)h
1
+···+X
1
(ω)h
n
)
P(dω) =
n
k=1
_
Ω
e
iX
k
(ω)h
k
P(dω)
= e
i(m(X
1
)h
1
+···+m(X
n
)h
n
)
e
−
1
2
(Q(X
1
)h
2
1
+···+Q(X
n
)h
2
n
)
.
Proposition 2.12 Assume that X
1
, ..., X
n
are real random variables and
that X = (X
1
, ..., X
n
) is Gaussian. Then X
1
, ..., X
n
are independent if and
only if Q(X) is diagonal.
Proof. If X
1
, ..., X
n
are independent the conclusion follows from Proposition
2.11. Assume now that Q(X) is diagonal. By Proposition 2.7 it is enough to
show that
X
#
P(h) =
n
i=1
(X
k
)
#
P(h),
for each h = (h
1
, ..., h
n
) ∈ H.
We have in fact
X
#
P(h) = e
im(X),h
e
−
1
2
Q(X)h,h
= e
im(X),h
e
−
1
2
P
n
k=1
Q(X)
k,k
h
2
k
= e
im(X),h
e
−
1
2
P
n
k=1
Q(X
k
)h
2
k
=
n
i=1
(X
k
)
#
P(h).
2.3 Gaussian random variables deﬁned in a
Hilbert space
We now consider the case when (Ω, F, P) coincides with (H, B(H), µ), where
H is a separable Hilbert space and µ = N
m,Q
with m ∈ H and Q ∈ L
+
1
(H).
22 Chapter 2
2.3.1 Aﬃne changes of variables
Let b ∈ K and A ∈ L(H, K) where K is another separable Hilbert space.
Let us consider the aﬃne transformation
T(x) = Ax + b, x ∈ H.
Proposition 2.13 T is a Gaussian random variable and its law T
#
µ is given
by N
Aa+b,AQA
∗, where A
∗
is the transpose of A.
Proof. We have in fact
_
K
e
ik,y
T
#
µ(dy) =
_
H
e
ik,T(x)
µ(dx) =
_
H
e
ik,Ax+b
µ(dx)
= e
ik,b
_
H
e
iA
∗
k,x
µ(dx) = e
ik,Aa+b
e
−
1
2
AQA
∗
k,k
, k ∈ K.
Example 2.14 Let µ = N
m,Q
and n ∈ N, f
1
, ..., f
n
∈ H. Let F : H → R
n
be deﬁned as
F(x) := (¸x, f
1
), ..., ¸x, f
n
)), x ∈ H.
Then by Proposition 2.13 F is a Gaussian random variable with mean m(F)
and covariance Q(F) given by,
m(F) = F(m) = (¸m, f
1
), ..., ¸m, f
n
))
and
Q(F) = FQF
∗
.
On the other hand, the linear operator F
∗
: R
n
→ H is given by
F
∗
(ξ) =
n
k=1
f
k
ξ
k
, ∀ ξ = (ξ
1
, ..., ξ
n
) ∈ R
n
.
Therefore
QF
∗
(ξ) =
n
k=1
Qf
k
ξ
k
, ∀ ξ = (ξ
1
, ..., ξ
n
) ∈ R
n
and
FQF
∗
(ξ) =
__
n
k=1
Qf
k
ξ
k
, f
1
_
, ...,
_
n
k=1
Qf
k
ξ
k
, f
n
__
random variables 23
so that
Q(F)
h,k
= ¸Qf
h
, f
k
). (2.2)
Therefore, F
1
, ..., F
n
are independent if and only if
¸Qf
h
, f
k
) = 0, h, k = 1, ..., n,
if h ,= k.
2.4 The white noise function
In order to deﬁne the white noise function (which will play an important role
in what follows), we shall deal with equivalence class of random variables
(rather than random variables), which we brieﬂy discuss in the next sub
section.
2.4.1 Equivalence classes of random variables
Let (Ω, F, P) be a probability space and let H be a separable Hilbert space.
We denote by R(H) the set of all Hvalued random variables.
Deﬁnition 2.15 We say that X, Y ∈ R(H) are equivalent (and write X ∼
Y ) if
P(¦ω ∈ Ω : X(ω) = Y (ω)¦) = 1.
One can easily check that X ∼ Y, X, Y ∈ R(H) is an equivalence relation,
so that the set R(H) is disjoint union of equivalences classes.
We notice that if X ∼ Y then the laws of X and Y coincide. In fact set
K = ¦ω ∈ Ω : X(ω) ,= Y (ω)¦,
so that P(K) = 0. Since for any I ∈ B(H) we have
X
−1
(I) ⊂ Y
−1
(I) ∪ K,
it follows that P(X
−1
(I)) ≤ P(Y
−1
(I)) and, exchanging X and Y we see that
P(X
−1
(I)) = P(Y
−1
(I)).
Consequently, all random variables belonging to a ﬁxed equivalence class
˜
X have the same law, which is called the law of
˜
X.
In the following we shall not distinguish between a random variable X
and the equivalence class
˜
X including X, except when needed.
24 Chapter 2
By L
p
(Ω, F, P; H), p ≥ 1, we mean the space of all equivalence class of
random variables X: Ω → H such that
_
Ω
[X(ω)[
p
P(dω) < +∞.
L
p
(Ω, F, P; H), endowed with the norm
X
L
p
(Ω,F,P;H)
=
__
Ω
[X(ω)[
p
P(dω)
_
1/p
,
is a Banach space. We shall write L
p
(Ω, F, P; H) = L
p
(Ω, P; H) for brevity.
We prove now that the limit of a convergent sequence in L
2
(Ω, P; H) of
Gaussian random variables is Gaussian.
Proposition 2.16 Let (X
n
) ⊂ L
2
(Ω, P; H) be a sequence of Gaussian ran
dom variables convergent to X in L
2
(Ω, P; H). Then X is a Gaussian random
variable and
¸m(X), h) = lim
n→∞
¸m(X
n
), h), h ∈ H,
and
¸Q(X)h, k) = lim
n→∞
¸Q(X
n
)h, k), h, k ∈ H.
Proof. Since X
n
→ X in L
2
(Ω, P; H) we have
lim
n→∞
¸m(X
n
), h) = lim
n→∞
_
Ω
¸X
n
(ω), h)P(dω) =
_
Ω
¸X(ω), h)P(dω) = ¸m(X), h)
and
lim
n→∞
¸Q(X
n
)h, k) = lim
n→∞
_
Ω
¸X
n
(ω) −m(X
n
), h) ¸X
n
(ω) −m(X
n
), k)P(dω)
=
_
Ω
¸X(ω) −m(X), h) ¸X(ω) −m(X), k)P(dω) = ¸Q(X)h, k).
Let us show now that X is a Gaussian random variable. We have in fact
_
H
e
ix,h
(X
#
µ)P(dy) =
_
Ω
e
iX(ω),h
P(dω) = lim
n→∞
_
Ω
e
iX
n
(ω),h
P(dω)
= lim
n→∞
e
im(X
n
),h
e
−
1
2
Q(X
n
)h,h
= e
im(X),k
e
−
1
2
Q(X)h,h
.
random variables 25
2.4.2 Deﬁnition of the white noise function
In this section we assume that the Hilbert space H is inﬁnite dimensional and
consider a non degenerate Gaussian measure µ = N
Q
in H (Ker (Q) = ¦0¦).
Since Q is compact there exists a complete orthonormal basis (e
k
) on H and
a sequence of positive numbers (λ
k
) such that
Qe
k
= λ
k
e
k
, k ∈ N.
Let us deﬁne a mapping
W : Q
1/2
(H) → C(H), z → W
z
where
W
z
(x) = ¸x, Q
−1/2
z), ∀ x ∈ H.
Here Q
1/2
(H) is the Cameron–Martin space and C(H) the space of all real
continuous functions on H.
Lemma 2.17 For all z
1
, z
2
∈ Q
1/2
(H) we have
_
H
W
z
1
(x)W
z
2
(x)µ(dx) = ¸z
1
, z
2
). (2.3)
Proof. We have in fact
_
H
W
z
1
(x)W
z
2
(x)µ(dx) =
_
H
¸x, Q
−1/2
z
1
)¸x, Q
−1/2
z
2
)µ(dx)
= ¸QQ
−1/2
z
1
, QQ
−1/2
z
2
) = ¸z
1
, z
2
).
Since Q
1/2
(H) is dense in H, the mapping W can be uniquely extended
as a mapping from H into L
2
(H, µ) which we denote still by W and call the
white noise function.
W
f
is linear in the sense that for all α, β ∈ R we have
W
f
(αx + βy) = αW
f
(x) + βW
f
(y), x, y µ a.e..
Remark 2.18 Given z ∈ H (not belonging to Q
1/2
(H)) it would be tempt
ing to deﬁne the random variable W
z
by setting,
W
z
(x) = ¸Q
−1/2
x, z), x ∈ Q
1/2
(H).
However this deﬁnition is meaningless because µ(Q
1/2
(H)) = 0, by Proposi
tion 1.13
26 Chapter 2
Proposition 2.19 Let z ∈ H. Then W
z
is a real Gaussian random variable
with mean 0 and covariance [z[
2
.
Proof. We have to show that
_
H
e
iηW
z
(x)
µ(dx) = e
−
1
2
η
2
z
2
, ∀ η ∈ R.
Let (z
n
) ⊂ Q
1/2
(H) be a sequence such that z
n
→ z in H. Then, by the
dominated convergence theorem, we have
_
H
e
iηW
z
(x)
µ(dx) = lim
n→∞
_
H
e
iηQ
−1/2
z
n
,x
µ(dx) = lim
n→∞
e
−
1
2
η
2
z
n

2
= e
−
1
2
η
2
z
2
.
So, the conclusion follows.
The following generalization of Proposition 2.19 is important.
Proposition 2.20 Let n ∈ N, z
1
, ..., z
n
∈ H. Then (W
z
1
, ..., W
z
n
) is an n
dimensional Gaussian random variable with mean 0 and covariance operator
Q
z
given by
(Q
z
)
h,k
= ¸z
h
, z
k
), h, k = 1, ..., n. (2.4)
The random variables W
z
1
, ..., W
z
n
are independent if and only if z
1
, ..., z
n
are
mutually orthogonal.
Proof. Let (z
1
j
), ..., (z
n
j
) be n sequences in Q
1/2
(H) convergent respectively to
z
1
, ..., z
n
in H. Then we have by the dominated convergence theorem, that
_
H
e
i(ξ
1
W
z
1
(x)+···+ξ
n
W
z
n
(x))
µ(dx) = lim
j→∞
_
H
e
i(ξ
1
Q
−1/2
z
j
1
,x+···+ξ
n
Q
−1/2
z
j
n
,x)
µ(dx)
= lim
j→∞
_
H
e
ix,Q
−1/2
(ξ
1
z
j
1
+···+ξ
n
z
j
n
)
µ(dx)
= lim
j→∞
e
−
1
2
ξ
1
z
j
1
+···+ξ
n
z
j
n

2
= e
−
1
2
ξ
1
z
1
+···+ξ
n
z
n

2
= e
−
1
2
P
n
j,k=1
z
j
,z
k
ξ
j
ξ
k
.
Chapter 3
Brownian Motion
3.1 Stochastic Processes
We are given a probability space (Ω, F, P). We denote by P
∗
the outer
measure of P. We recall that a null set of Ω is a set of outer measure zero.
For any integrable real random variable F we note
E(F) =
_
Ω
F(ω)P(dω).
So, in particular we have
F
#
P(I) = E(1l
I
(F)), ∀ I ∈ B(R).
We say that a property π concerning elements of Ω holds Pa.s. if the set
where π does not hold is a null set.
Deﬁnition 3.1 A family X = (X(t))
t≥0
of real random variables in (Ω, F, P)
is called a real stochastic process in [0, +∞). For any ω ∈ Ω, X(, ω) is called
a trajectory of X.
• X is Gaussian if for any n ∈ N and any 0 ≤ t
1
< < t
n
the n
dimensional random variable (X(t
1
), ..., X(t
n
)) is Gaussian.
• X is continuous if X(, ω) is continuous Pa.s.
• X is pmean continuous, p ≥ 1, if
(i) X(t) is pintegrable for any t ≥ 0.
(ii) We have
lim
t→t
0
E[[X(t) −X(t
0
)[
p
] = 0, ∀ t
0
≥ 0. (3.1)
27
28 Chapter 3
We notice that a pmean continuous process is not continuous in general.
We say that two stochastic processes X and Y are equivalent if for all
t ≥ 0 we have
X(t, ω) = Y (t, ω), Pa.s..
When X and Y are equivalent we also say that Y is a version of X (or that
X is a version of Y ).
3.2 Brownian motion
Deﬁnition 3.2 A real Brownian motion B = (B(t))
t≥0
on (Ω, F, P) is a
real stochastic process such that
(i) B(0) = 0 and if 0 ≤ s < t, B(t) − B(s) is a real Gaussian random
variable with law N
t−s
.
(ii) If 0 < t
1
< ... < t
n
, the random variables,
B(t
1
), B(t
2
) −B(t
1
), , B(t
n
) −B(t
n−1
)
are independent.
We express condition (ii) by saying that B is a process with independent
increments.
Lemma 3.3 Let t, s > 0. Then
E[B(t)(B(s)] = min¦t, s¦. (3.2)
Proof. Let for instance t > s. Then we have
E[B(t)B(s)] = E[(B(t) −B(s))B(s)] +E[B
2
(s)].
On the other hand, B(t) −B(s) is independent of B(s) so that
E[(B(t) −B(s))B(s)] = E[B(t) −B(s)]E[B(s)] = 0.
Since the law of B(s) is N
s
we conclude that E[B(t)B(s)] = s as required.
Brownian motion 29
3.2.1 Construction of a Brownian motion
Consider the probability space (H, B(H), µ), where H = L
2
(0, +∞) and
µ = N
Q
, Q being an arbitrary (but ﬁxed) non degenerate Gaussian measure
in H.
Deﬁne
B(t) = W
1l
[0,t]
, t ≥ 0, (3.3)
where
1l
[0,t]
(s) =
_
_
_
1 if s ∈ [0, t],
0 otherwise,
and W is the white noise function deﬁned in Chapter 2.
More precisely, for any t ≥ 0 we choose an arbitrary element in the
equivalence class of B(t) which we still denote by B(t).
Clearly, for any t ≥ 0, B(t) is a Gaussian random variable N
t
and for any
t > s ≥ 0, B(t) −B(s) = W
1l
(s,t]
is a Gaussian random variable N
t−s
. So, B
fulﬁlls Deﬁnition 3.2(i). Let us prove (ii). Since the system of elements of H,
(1l
[0,t
1
]
, 1l
(t
1
,t
2
]
, ..., 1l
(t
n−1
,t
n
]
),
is orthogonal, we have by Proposition 2.20 that the random variables
B(t
1
), B(t
2
) −B(t
1
), , B(t
n
) −B(t
n−1
)
are independent. Thus (ii) is proved as well.
3.2.2 Some properties of a Brownian motion
Proposition 3.4 Let B(t), t ≥ 0, be a Brownian motion on (Ω, F, P).
Then B is a Gaussian process. Moreover, if 0 < t
1
< ... < t
n
the law of
(B(t
1
), ..., B(t
n
)) is given by
P((B(t
1
), ..., B(t
n
)) ∈ I)
= (2π)
−n/2
(t
1
(t
2
−t
1
) (t
n
−t
n−1
))
−1/2
_
I
e
−
η
2
1
2t
1
−
(η
2
−η
1
)
2
2(t
2
−t
1
)
−·−
(η
n
−η
n−1
)
2
2(t
n
−t
n−1
)
dη,
(3.4)
for all I ∈ B(R
n
).
Proof. Let 0 < t
1
< ... < t
n
and set
X := (B(t
1
), B(t
2
) −B(t
1
), ..., B(t
n
) −B(t
n−1
))
Z := (B(t
1
), ..., B(t
n
)).
30 Chapter 3
Since random variables B(t
1
), B(t
2
) − B(t
1
), ..., B(t
n
) − B(t
n−1
) are inde
pendent, by Proposition 2.11 it follows that X is a ndimensional Gaussian
random variable with mean 0 and covariance operator
Q(X) = diag (t
1
, t
2
−t
1
, ..., t
n
−t
n−1
).
Now, consider the linear mapping T ∈ L(R
n
) deﬁned by,
T(x
1
, ..., x
n
) = (x
1
, x
1
+ x
2
, ..., x
1
+ + x
n
), ∀ (x
1
, ..., x
n
) ∈ R
n
.
It is clear that Z = T(X). Therefore by Proposition 2.13 Z is Gaussian with
mean 0 and covariance Q(Z) = TQ(X)T
∗
where T
∗
is the transpose of T.
It remain to show (3.4). If I ∈ B(R
n
) we have
P(Z ∈ I) = (2π)
−n/2
(det Q(Z))
−1/2
_
I
e
−
1
2
(Q(Z))
−1
η,η
dη.
Since det T = det T
∗
= 1, as easily checked, we have
det Q(Z) = det Q(X) = t
1
(t
2
−t
1
) (t
n
−t
n−1
).
Moreover, since
T
−1
η = (η
1
, η
2
−η
1
, ..., η
n
−η
n−1
),
we have
¸(Q(Z))
−1
η, η) = ¸Q
−1
T
−1
η, T
−1
η) =
η
2
1
t
1
−
(η
2
−η
1
)
2
(t
2
−t
1
)
− −
(η
n
−η
n−1
)
2
(t
n
−t
n−1
)
and so, the conclusion follows.
Proposition 3.5 Let B(t), t ≥ 0, be a Brownian motion on (Ω, F, P).
Then B is pmean square continuous for all p ≥ 1.
Proof. It is enough to show the result for p = 2m, m ∈ N. Let t > t
0
≥ 0.
Since B(t) −B(t
0
) is a Gaussian random variable N
t−t
0
, we have
E([B(t) −B(t
0
)[
2m
) =
_
R
[ξ[
2m
N
t−t
0
(dξ) =
(2m)!
m!2
m
(t −t
0
)
m
.
Therefore
lim
t→0
E([B(t) −B(t
0
)[
2m
) = 0
and the conclusion follows.
Exercise 3.6 Let B(t) be a Brownian motion in a probability space (Ω, F, P).
Prove that the following are Brownian motions.
Brownian motion 31
(i) B
1
(t) = B(t + h) −B(h), t ≥ 0, where h > 0 is given.
(ii) B
2
(t) = αB(α
−2
t), t ≥ 0, where α > 0 is given.
(iii) B
3
(t) = tB(1/t), t > 0, B
3
(0) = 0.
(iv) B
4
(t) = −B(t), t ≥ 0.
3.3 Wiener integral
Let B(t), t ≥ 0, be a Brownian motion in (Ω, F, P) and let f ∈ L
2
(0, T)
with T > 0. We want to deﬁne the stochastic integral:
_
T
0
f(s)dB(s).
We start with step functions. Let 0 = t
0
< t
1
< < t
n
= T, f
0
, f
1
, ..., f
n−1
∈
R and set
f =
n
j=1
t
j−1
1l
(t
j
−t
j−1
]
.
Then deﬁne
_
T
0
f(s)dB(s) :=
n
j=1
f
t
j−1
(B(t
j
) −B(t
j−1
)).
Let us prove two basic identities.
Lemma 3.7 We have
E
__
T
0
f(s)dB(s)
_
= 0 (3.5)
and
E
_
__
T
0
f(s)dB(s)
_
2
_
=
n
j=1
[f(t
j−1
)[
2
(t
j
−t
j−1
) =
_
t
0
f
2
(s)ds. (3.6)
Proof. Identity (3.5) is obvious. Let us prove (3.6). We have
E([I
σ
(f)[
2
) = E
_
n
j=1
[f(t
j−1
)[
2
[B(t
j
) −B(t
j−1
)]
2
_
+2E
_
n
j<k
f(t
j−1
)f(t
k−1
)[B(t
j
) −B(t
j−1
)][B(t
k
) −B(t
k−1
)]
_
.
(3.7)
32 Chapter 3
Now the conclusion follows taking into account that B(t
j
) −B(t
j−1
) is a real
Gaussian random variable N
t
j−1
−t
j
and that B(t
j
) − B(t
j−1
) is independent
of B(t
k
) −B(t
k−1
) for k ,= j.
Denote by S(0, T) the linear space of all step functions. By (3.6) it follows
that the linear mapping I
S(0, T) ⊂ L
2
(0, T) → L
2
(Ω, F, P), f → I(f) =
_
T
0
f(s)dB(s),
is continuous. Since S(0, T) is dense in L
2
(0, T) it can be uniquely extended
to the whole L
2
(0, T). We still denote by I(f) =
_
T
0
f(s)dB(s) this estension.
It is clear that for any f ∈ L
2
(0, T) we have
E
__
T
0
f(s)dB(s)
_
= 0, (3.8)
and
E
_
__
T
0
f(s)dB(s)
_
2
_
=
_
t
0
f
2
(s)ds. (3.9)
The random variable (more precisely, the equivalence class of random
variables)
_
T
0
f(s)dB(s), which belongs to L
2
(Ω, F, P), is called the Wiener
integral of f in [0, T].
We deﬁne in an obvious way the Wiener integral
_
b
a
f(s)dB(s) for any
a, b ≥ 0. It is easy to see that if a, b, c ≥ 0 we have
_
b
a
f(s)dB(s) +
_
c
b
f(s)dB(s) =
_
c
a
f(s)dB(s).
Exercise 3.8 Let f, g ∈ L
2
(0, T). Show that
E
__
T
0
f(s)dB(s)
_
T
0
g(s)dB(s)
_
=
_
T
0
f(s)g(s)ds.
Proposition 3.9 Let f ∈ L
2
(0, T). Then I(f) =
_
T
0
f(s)dB(s) is a real
Gaussian random variable N
q
with q =
_
T
0
[f(s)[
2
ds.
Proof. It is enough to prove the result for f of the form
f =
n
i=1
f
t
i−1
(t
i
−t
i−1
),
Brownian motion 33
where n ∈ N, 0 = t
0
< t
1
< ... < t
n−1
= T, so that
I(f) =
n
i=1
f
t
i−1
(B(t
i
) −B(t
i−1
)).
Since random variables
B(t
1
), B(t
2
) −B(t
1
), , B(t
n
) −B(t
n−1
),
are independent, we have that I(f) is a real Gaussian random variable N
q
with
q =
n
i=1
f
2
(t
i−1
)(t
i
−t
i−1
).
We now show a relation between the white noise function and the Wiener
integral.
Example 3.10 We use here notations of Section 3.2.1. Let f ∈ L
2
(0, ∞).
Then we have
W
f
=
_
∞
0
f(s)dB(s). (3.10)
It is enough to show (3.10) when
f =
n
k=1
f
t
k−1
1l
(t
k−1
,t
k
]
,
where 0 ≤ t
0
< < t
n
. In this case we have in fact
_
∞
0
f(s)dB(s) =
n
k=1
f
t
k−1
W
1l
(t
k−1
,t
k
]
= W
P
n
k=1
f
t
k−1
1l
(t
k−1
,t
k
]
= W
f
.
Let f : [0, ∞) →R such that it is integrable in all interval [0, T], T0. Let
us introduce a stochastic process setting
F(t) =
_
t
0
f(s)ds, ∀ t ≥ 0.
Proposition 3.11 The process F(t), t ≥ 0 is pmean continuous for any
p ≥ 1.
34 Chapter 3
Proof. Let p = 2m, m ∈ N and t > t
0
≥ 0. Then by Proposition 3.9 we
have that
F(t) −F(t
0
) =
_
t
t
0
f(s)dB(s)
is a real Gaussian random variable with mean 0 and covariance
_
t
t
0
f
2
(s)ds.
Therefore
E[F(t) −F(t
0
)[
2m
=
(2m)!
m!2
m
__
t
t
0
f
2
(s)ds
_
q
,
so that
lim
t→t
0
E[F(t) −F(t
0
)[
2m
= 0.
We note ﬁnally, that if f ∈ C
1
([0, T]) then it is possible to express the
Wiener integral
_
T
0
f(s)dB(s) in terms of a Riemann integral as the following
integration by parts formula shows.
Proposition 3.12 If f ∈ C
1
([0, T]) we have
_
T
0
f(s)dB(s) = f(T)B(T) −
_
T
0
f
(s)B(s)ds, Pa.e. ω ∈ Ω. (3.11)
Proof. Let σ = ¦t
0
, t
1
, , t
n
¦ ∈ Σ. Then we have
I
σ
(f) =
n
k=1
f(t
k−1
)(B(t
k
) −B(t
k−1
))
=
n
k=1
(f(t
k
)B(t
k
) −f(t
k−1
)B(t
k−1
))
−
n
k=1
(f(t
k
) −f(t
k−1
))B(t
k
)
= f(T)B(T) −
n
k=1
(f(t
k
) −f(t
k−1
))B(t
k
)
= f(T)B(T) −
n
k=1
f
(α
k
)B(t
k
)(t
k
−t
k−1
),
where α
k
are suitable numbers in the interval [t
k−1
, t
k
], k = 1, ..., n. It follows
that
lim
σ→0
I
σ
(f) = f(T)B(T) −
_
T
0
f
(s)dB(s)ds, Pa.s..
Brownian motion 35
3.4 Continuity of Brownian motion
Let B(t), t ≥ 0, be a Brownian motion on a probability space (Ω, F, P). We
are going to show that B possesses a continuous version. To this purpose we
shall use a representation formula for B proved in the next proposition.
Proposition 3.13 For any α ∈ (0, 1/2) we have
B(t) =
sin πα
π
_
t
0
(t −σ)
α−1
Y
α
(σ)dσ, (3.12)
where
Y
α
(σ) =
_
σ
0
(σ −s)
−α
dB(s). (3.13)
Notice that the Wiener integral Y
α
is meaningful since α ∈ (0, 1/2).
Proof. We start from the following elementary identity which is valid for
any α ∈ (0, 1).
_
t
s
(t −σ)
α−1
(σ −s)
−α
dσ =
π
sin πα
, 0 ≤ s ≤ σ ≤ t, (3.14)
where α ∈ (0, 1). To check (3.14) it is enough to set σ = r(t −s) +s so that
(3.14) becomes
_
1
0
(1 −r)
α−1
r
−α
dr = β(α, 1 −α) =
π
sin πα
.
Now since, obviously, B(t) =
_
s
0
dB(s) we can write
B(t) =
sin πα
π
_
t
0
__
t
s
(t −σ)
α−1
(σ −s)
−α
dσ
_
dB(s).
Exchanging integrals
(1)
, yields
B(t) =
sin πα
π
_
t
0
dξ(t −σ)
α−1
__
σ
0
(σ −s)
−α
dB(s)
_
.
We can now prove the result.
Theorem 3.14 Let B(t), t ≥ 0, be a Brownian motion on a probability
space (Ω, F, P). Then B possesses a continuous version.
(1)
This requires a proof which is left to the reader.
36 Chapter 3
Proof. Choose a version Y
α
(, ω) of the stochastic process Y
α
which is 2m
integrable with 2m > 1/α. This is possible in view of Proposition 3.11. Now
set
B(t, ω) =
sin πα
π
_
t
0
(t −σ)
α−1
Y
α
(σ, ω)dσ, ∀ t ≥ 0.
Then B(, ω) is a continuous version of B thanks to the following analytic
lemma.
Lemma 3.15 Let α ∈ (0, 1/2), m ∈ N with 2m > 1/α and f ∈ L
2m
(0, T).
Set
F(t) =
_
t
0
(t −σ)
α−1
f(σ)dσ, t ∈ [0, T].
Then F ∈ C([0, T]; H).
Proof. By H¨ older’s inequality we have
[F(t)[ ≤
__
t
0
(t −σ)
(α−1)
2m
2m−1
dσ
_
2m−1
2m
[f[
L
2m
(0,T;H)
. (3.15)
(Notice that (α − 1)
2m
2m−1
> −1.) Therefore F ∈ L
∞
(0, T; H) and F is con
tinuous at 0. Let us prove that F is continuous on [
t
0
2
, T] for any t
0
∈ (0, T].
Let us set for ε <
t
0
2
,
F
ε
(t) =
_
t−ε
0
(t −σ)
α−1
f(σ)dσ, t ∈ [0, T].
F
ε
is obviously continuous on [
t
0
2
, 1]. Moreover, using again H¨ older’s inequal
ity, we ﬁnd
[F(t) −F
ε
(t)[ ≤ M
_
2m−1
2mα −1
_2m−1
2m
ε
α−
1
2m
[f[
L
2m
(0,T;H)
.
Thus lim
ε→0
F
ε
(t) = F(t), uniformly on [
t
0
2
, T], and F is continuous as re
quired.
Exercise 3.16 Prove that B possesses an H¨ older continuous version with
any exponent β < 1/2.
3.5 The standard Brownian motion
Let us consider a Brownian motion B(t), t ≥ 0, in a probability space
(Ω, F, P) such that B(, ω) is continuous for all ω ∈ Ω. We denote by B
the mapping
B : Ω → C
0
, ω → B(, ω),
where C
0
= ¦η ∈ C([0, +∞)) : η(0) = 0¦.
Brownian motion 37
3.5.1 Some properties of C
0
First we notice that, as easily checked, C
0
, endowed with the metric,
d(η
1
, η
2
) :=
∞
k=1
η
1
−η
2

k
2
k
(1 +η
1
−η
2

k
)
,
is a complete metric space. We have set for any k ∈ N,
η
k
= sup¦[η(t)[ : t ∈ [0, k]¦, ∀ η ∈ C
0
.
Let us now consider the σalgebra B(C
0
). It is important to notice that
B(C
0
) is generated by the cylindrical subsets of C
0
that we shall introduce
now.
For n ∈ N, 0 < t
1
< < t
n
and A ∈ B(R
n
) we deﬁne
C
t
1
,t
2
,...,t
n
;A
:= ¦η ∈ C
0
: (η(t
1
), ..., η(t
n
)) ∈ A¦ .
Note that
C
t
1
,t
2
,...,t
n
;A
= C
t
1
,t
2
,...,t
n
,t
n+1
,...,t
n+k
;A×R
k, k, n ∈ N.
Using this identity one can easily see that C is an algebra. Moreover, the
σalgebra generated by C coincides with B(C
0
) since any ball (with respect
to the metric of C
0
) is a countable intersection of cylindrical sets.
3.5.2 The Wiener measure and the standard Brownian
motion
We come back to the mapping B
B : Ω → C
0
, ω → B(, ω)
and we denote by Q its law (which is a probability measure on (C
0
, B(C
0
)).
Q is called the Wiener measure on (C
0
, B(C
0
)).
So, for any nonnegative Borel mapping
F : C
0
→R, η → F(η),
we have
E[F(B())] =
_
Ω
F(B(, ω))P(dω) =
_
C
0
F(η)Q(dη). (3.16)
Some examples of mappings F are the following.
38 Chapter 3
(i) F(η) = g(η(t
0
)), for all η ∈ C
0
, where g : R → R is nonnegative Borel
and t
0
> 0 is given.
(ii) F(η) = G(η(t
1
), ..., η(t
n
)), for all η ∈ C
0
, where G : R
n
→R is nonneg
ative Borel and t
1
, ..., t
n
> 0 are given.
(iii) F(η) = sup
t∈[0,1]
[η(t)[, for all η ∈ C
0
.
Now we deﬁne a stochastic process W(t), t ≥ 0, in (C
0
, B(C
0
), Q) setting
W(t)(η) = η(t), η ∈ C
0
, t ≥ 0.
Proposition 3.17 W is a Brownian motion in (C
0
, B(C
0
), Q), called the
standard Brownian motion.
Proof. The proof is straightforward. Let us show for instance that for
t > s ≥ 0, W(t) − W(s) is a Gaussian random variable N
t−s
. For this it is
enough to show that the Fourier transform of W(t) −W(s)
ψ(h) :=
_
C
0
e
i(η(t)−η(s))h
Q(dη), h ∈ R,
is given by e
−
1
2
(t−s)h
2
, h ∈ R.
In fact by (3.16) we have
_
C
0
e
i(η(t)−η(s))h
Q(dη) =
_
Ω
e
i(B(t,ω)−B(s,ω))h
P(dω)
= E[e
i(B(t)−B(s))
] = e
−
1
2
(t−s)h
2
, h ∈ R.
In an analogous way one can prove that W(t), t ≥ 0, has independent incre
ments.
Let us compute the Wiener measure of a cylindrical set.
Proposition 3.18 Let C
t
1
,t
2
,...,t
n
;A
be a cylindrical set. Then we have
Q(C
t
1
,t
2
,...,t
n
;A
)
=
1
_
(2π)
n
t
1
(t
2
−t
1
) (t
n
−t
n−1
)
_
A
e
−
ξ
2
1
2t
1
−
(ξ
2
−ξ
1
)
2
2(t
2
−t
1
)
−···−
(ξ
n
−ξ
n−1
)
2
2(t
n
−t
n−1
)
dξ.
Proof. We simply note that, thanks to (3.16), we have
Q(C
t
1
,t
2
,...,t
n
;A
) = P((B(t
1
), ..., B(t
n
)) ∈ A),
so that the conclusion follows from Proposition 3.4.
Brownian motion 39
3.6 Quadratic variation of the Brownian mo
tion
In this section we are given a real continuous Brownian motion B(t), t ≥ 0,
on a probability space (Ω, F, P). For any T > 0 we denote by Σ(0, T) the
set of all decompositions of [0, T]
σ = ¦0 = t
0
< t
1
< < t
n
= T¦.
Then for any σ = ¦0 = t
0
< t
1
< < t
n
= T¦ ∈ Σ(0, T) we set
[σ[ := min¦t
k
−t
k−1
: k = 1, ...n −1¦.
We introduce a partial ordering on Σ(0, T), setting
σ
1
≤ σ
2
if and only if [σ
1
[ ≤ [σ
2
[.
Let us now introduce the quadratic variation of Brownian motion B in
[0, T]. For any σ = ¦0 = t
0
< t
1
< < t
n
= T¦ ∈ Σ(0, T) we deﬁne
J
σ
:=
n
k=1
[B(t
k
) −B(t
k−1
)[
2
.
Then we prove
Theorem 3.19 We have
lim
σ→0
J
σ
= T in L
2
(Ω, F, P).
We say that T is the quadratic variation of B in [0, T].
Proof. Since B
t
k
−B
t
k−1
is a real Gaussian random variable with law N
t
k
−t
k−1
,
we have E(J
σ
) = T, and so,
E([J
σ
−T[
2
) = E(J
2
σ
) −2TE(J
σ
) + T
2
= E(J
2
σ
) −T
2
. (3.17)
Moreover
E[J
σ
[
2
= E
¸
¸
¸
¸
¸
n
k=1
[B(t
k
) −B(t
k−1
)[
2
¸
¸
¸
¸
¸
2
= E
n
k=1
[B(t
k
) −B(t
k−1
)[
4
+ 2
n
h<k=1
E[B(t
h
) −B(t
h−1
)[
2
[B(t
k
) −B(t
k−1
)[
2
.
40 Chapter 3
But we have
E
n
k=1
[B(t
k
) −B(t
k−1
)[
4
= 3
n
k=1
(t
k
−t
k−1
)
2
, (3.18)
and, since B(t
h
) −B(t
h−1
) and B(t
k
) −B(t
k−1
) are independent, we have
n
h<k=1
E[B(t
h
) −B(t
h−1
)[
2
[B(t
k
) −B(t
k−1
)[
2
=
n
h<k=1
(t
h
−t
h−1
)(t
k
−t
k−1
).
(3.19)
Therefore
E[J
σ
[
2
= 3
n
k=1
(t
k
−t
k−1
)
2
+ 2
n
h<k=1
(t
h
−t
h−1
)(t
k
−t
k−1
)
= 2
n
k=1
(t
k
−t
k−1
)
2
+
_
n
k=1
(t
k
−t
k−1
)
_
2
.
= 2
n
k=1
(t
k
−t
k−1
)
2
+ T
2
.
(3.20)
Now, substituting (3.20) on (3.17), we obtain
E
_
[J
σ
−T[
2
_
= 2
n
k=1
(t
k
−t
k−1
)
2
→ 0,
as [σ[ → 0.
An important consequence of Theorem 3.19 is that almost all trajectories
of the Brownian motion B have not bounded variation
(2)
. In other terms
the set
V
T
:= ¦ω ∈ Ω : B(, ω) ∈ BV (0, T)¦
has outer probability zero.
In fact the following result holds.
Proposition 3.20 We have P
∗
(V
T
) = 0.
(2)
Let f : [0, T] → R. Then for any σ = ¦0 = t
0
< t
1
< < ..., t
n
= T¦ ∈ Σ(0, T) we
set V
σ
(f) =
n
k=1
[f(t
k
) − f(t
k−1
)[ and deﬁne V (f) := sup
σ∈Σ
V
σ
(f), V (f) is called the
variation of f. BV (0, T) is the set of all functions f : [0, T] →R of ﬁnite variation.
Brownian motion 41
Proof. Set
Λ := ¦ω ∈ Ω : B(, ω) is continuous ¦,
so that P(Λ) = 1 because B is continuous.
Since lim
σ→0
J
σ
= T in L
2
(Ω, F, P) there exists a sequence (σ
n
) ⊂
Σ(0, T) such that [σ
n
[ → 0 and a set Λ
1
⊂ F such that
(i) P(Λ
1
) = 1.
(ii) lim
n→∞
J
σ
n
(ω) = T for all ω ∈ Λ
1
.
We claim that
V
T
∩ Λ ⊂ Λ
c
1
. (3.21)
By the claim the conclusion will follow since P(Λ
c
1
) = 0.
Let us prove the claim. Let ω ∈ V
T
∩ Λ. Since B(, ω) is uniformly
continuous in [0, T], for any ε > 0 there exists δ
ε
> 0 such that
t, s ∈ [0, T], [t −s[ < δ
ε
=⇒ [B(t, ω) −B(s, ω)[ < ε.
Consequently, if n is so large that [σ
n
[ < δ
ε
we have J
σ
n
(ω) ≤ εV (B(, ω)).
Since ε is arbitrary ω cannot belong to Λ
1
. The claim is proved.
3.7 Multidimensional Brownian motions
Deﬁnition 3.21 Let n ∈ N and let X
1
, ..., X
n
be stochastic processes on a
probability space (Ω, F, P). Then X(t) := (X
1
(t), ..., X
n
(t)), t ≥ 0, is called
an ndimensional stochastic process.
X
1
, ..., X
n
are said to be independent if for any t
1
, ..., t
n
∈ [0, +∞) the
random variables X
i
(t
i
) are independent.
A ndimensional Brownian motion is a ndimensional stochastic process
B(t) := (B
1
(t), ..., B
n
(t)), t ≥ 0, such that B
1
, ..., B
n
are independent Brow
nian motions.
Example 3.22 Let us construct an ndimensional Brownian motion. Let
(e
1
, ..., e
n
) be the canonical basis in R
n
and choose Ω = H = L
2
(0, +∞; R
n
),
F = B(H) and P = N
Q
, where Q is any operator in L
+
1
(H) such that Ker
Q = ¦0¦.
Then set
B
i
(t) = W
e
i
1l
[0,t]
, ∀ t ≥ 0, i = 1, ..., n.
Then one can check easily that B(t) = (B
1
(t), ..., B
n
(t)) is an ndimensional
Brownian motion.
42 Chapter 3
Let B be a Brownian motion in R
n
. Then the following properties are
easily checked.
(i) If t > s, B(t) −B(s) is a Gaussian random variable with law N
(t−s)I
n
,
t ≥ 0, where I
n
represents the identity in R
n
,
(ii) E[B
i
(t)B
j
(t)] = 0 if i ,= j.
(iii) We have
E
_
[B(t) −B(s)[
2
¸
= n(t −s). (3.22)
Let us check (iii). We have
E
_
[B(t) −B(s)[
2
¸
=
n
k=1
E
_
[B
k
(t) −B
k
(s)[
2
¸
= n(t −s).
Exercise 3.23 Prove that for 0 ≤ s < t we have
E
_
[B(t) −B(s)[
4
¸
= (2n + n
2
)(t −s)
2
. (3.23)
Exercise 3.24 Let A, C ∈ L(R
d
) and set
Z(t) = e
tA
x +
_
t
0
e
(t−s)A
CdB(s), t ≥ 0.
Prove that the law of Z(t) in R
d
is given by
N
e
tA
x,Q
t
, (3.24)
where
Q
t
=
_
t
0
e
sA
CC
∗
e
sA
∗
ds, (3.25)
where A
∗
and C
∗
are the adjoint of A and C respectively.
Chapter 4
Markov property of the
Brownian motion
Let us consider the probability space (C
0
, B(C
0
), Q) where C
0
is the complete
metric space of all continuous functions ω : [0, +∞) → R introduced in
Chapter 3 and Q is the Wiener measure. Moreover, let W(t), t ≥ 0, the
standard Brownian motion in (C
0
, B(C
0
), Q) deﬁned by
W(t)(ω) = ω(t), ∀ t ≥ 0, ω ∈ C
0
.
This chapter is devoted to some sharp properties of the Brownian motion,
in particular the Markov and strong Markov property and the reﬂexion prin
ciple. To this purpose we shall introduce some basic concepts as ﬁltration,
stopping time and transition semigroup.
4.1 Filtration
For any t > 0 we denote by C
t
the algebra of all cylindrical sets
C
t
1
,··· ,t
n
;A
= ¦ω ∈ C
0
: (ω(t
1
), ..., ω(t
n
)) ∈ A¦
= ¦ω ∈ C
0
: (W(t
1
), ..., W(t
n
)) ∈ A¦
where 0 ≤ t
1
< ... < t
n
, t
n
≤ t and A ∈ B(R
n
). Moreover, we denote by F
t
the σalgebra generated by C
t
. Obviously F
0
= ¦∅, Ω¦.
The family of σ–algebras (F
t
)
t≥0
is increasing; it is called the natural
ﬁltration of W. For any t > 0 we deﬁne
F
t
− = σ¦F
t−
: ∈ (0, t)¦
43
44 Chapter 4
where σ
_
∈(0,t)
F
t−
_
is the σalgebra generated by F
t−
for ∈ (0, t) and
F
t
+ : =
>0
F
t+
, t ≥ 0.
Proposition 4.1 For all t > 0 we have F
t
= F
t
−.
Due to Proposition 4.1 we say that the natural ﬁltration (F
t
)
t≥0
is left con
tinuous.
Proof. Let t > 0. It is clear that
F
t
⊃
_
∈(0,t)
F
t−
,
so that F
t
⊃ F
t
−. To prove the converse inclusion it is enough to show that
C
t
⊂ F
t
−.
Let in fact I = C
t
1
,··· ,t
n
;A
∈ C
t
so that t
n
≤ t. If t
n
< t then I belongs to F
t
−
whereas if t
n
= t we have
I = lim
k→∞
C
t
1
,··· ,t
t−
1
k
;A
∈ F
t
−,
so that I ∈ F
t
− as well.
Remark 4.2 The ﬁltration (F
t
)
t≥0
is not right continuous, that is F
t
+ ,= F
t
for all t ≥ 0. Let for instance t = 0 and consider the sets
A
n
= ¦ω ∈ Ω : [ω(1/n)[ ≤ 1/n¦, n ∈ N.
Then A
n
∈ F
1/n
and A =
n∈N
A
n
∈ F
0
+. Notice that
A = ¦ω ∈ Ω : [ω
(0)[ = 0¦,
so that F
0
+ ,= F
0
.
4.1.1 F
t
measurable random variables
We say that a real random variable X is F
t
measurable if
I ∈ B(R) ⇒ X
−1
(I) ∈ F
t
.
In this case we say also that X depends from the story of the Brownian
motion only up to t.
The following lemma will be frequently used.
Markov property 45
Lemma 4.3 Let s
2
> s
1
≥ t > 0, and let ϕ be a real random variable
F
t
–measurable. Then W(s
2
) −W(s
1
) and ϕ are independent.
Proof. It is enough to show that for any A ∈ F
t
, W(s
2
) −W(s
1
) and 1l
A
are
independent; in other words that F
t
coincides with the set D deﬁned below.
D = ¦A ∈ F
t
: 1l
A
is independent of W(s
2
) −W(s
1
)¦.
Since W is a process with independent increments, D contains the algebra
of all cylindrical set belonging to C
t
(which is a πsystem). Moreover, D
is a λsystem. In fact if A ∈ D it is obvious that A
c
∈ D. Moreover, if
(A
n
) is a sequence in D consisting of disjoint sets, one can show easily that
∞
n=1
A
n
∈ D. Now the claim follows from Dynkin’s theorem (Theorem A.1
in Appendix A).
Next result shows that F
0
+ contains only trivial sets.
Proposition 4.4 (onezero law) Assume that A ∈ F
0
+. Then either P(A) =
1 or P(A) = 0.
Proof. Let A ∈ F
0
+. Denote by G the σalgebra generated by all sets of
the form
D
t
1
,...,t
n
,h;I
= ¦ω ∈ Ω : (ω(t
1
+ h) −ω(h), ..., ω(t
n
+ h) −ω(h)) ∈ I¦,
where n ∈ N, 0 < t
1
< < t
n
, h > 0, I ∈ B(R
n
). It is clear that A is
independent of G, since it belongs to all F
t
, t > 0, and W has independent
increments. Then we have
P(A ∩ G) = P(A)P(G), ∀ G ∈ G. (4.1)
On the other hand, we claim that G = B(C
0
). To prove the claim it is
enough to show that any cylindrical set C
t
1
,...,t
n
,h;I
belongs to G; but this
follows from the identity
lim
j→∞
D
t
1
−
1
j
,...,t
n
−
1
j
,
1
j
;I
= lim
j→∞
¦ω ∈ Ω : (ω(t
1
) −ω(1/j), ..., ω(t
n
) −ω(1/j)) ∈ I¦ = C
t
1
,...,t
n
;I
.
Since G = B(C
0
) we can set in (4.1) G = A, so that P
2
(A) = P(A) which
yields P(A) equal to zero or one.
Remark 4.5 For any t ≥ 0 denote by F
t
the σalgebra generated by F
t
and all null sets of Ω (called the completion of F
t
). By using Proposition 4.4
one can easily show that (F
t
)
t≥0
is both right and left continuous.
46 Chapter 4
4.2 Stopping times
A nonnegative extended (that is with values in [0, +∞]) random variable τ in
(C
0
, B(C
0
), Q) is called a stopping time with respect to the ﬁltration (F
t
)
t≥0
if
¦τ ≤ t¦ ∈ F
t
for all t ≥ 0.
To any stopping time τ we associate the σalgebra
F
τ
: = ¦A ∈ F : A ∩ ¦τ ≤ t¦ ∈ F
t
for all t ≥ 0¦.
Let us describe the σalgebra F
τ
, For 0 < t
1
< ... < t
n
and I∈B(R) we
deﬁne
C
(τ)
t
1
,...,t
n
;I
= ¦ω ∈ Ω : t
n
(ω) < τ, (ω(t
1
), ..., ω(t
n
)) ∈ I¦ = C
t
1
,...,t
n
;I
∩¦t
n
< τ¦.
We claim that C
(τ)
t
1
,...,t
n
;I
is F
τ
measurable.
In fact
C
(τ)
t
1
,...,t
n
;I
∩ ¦τ ≤ t¦ = C
t
1
,...,t
n
;I
∩ ¦t
n
< τ ≤ t¦
So, the σalgebra generated by all C
(τ)
t
1
,...,t
n
;I
in included in F
τ
and one can
show that it coincides with F
τ
.
If τ is stopping time, then ¦τ > t¦ and ¦τ = t¦ belong obviously to F
t
for all t ≥ 0.
Moreover, τ is F
τ
measurable. In fact, if A = ¦τ ≤ s¦ we have
A ∩ ¦τ ≤ t¦ = ¦τ ≤ t ∧ s¦ ∈ F
t∧s
⊂ F
t
.
In other words we have
F
τ
⊃ σ(τ),
where σ(τ) is the σalgebra generated by τ.
Remark 4.6 Let τ be an extended random variable such that
¦τ < t¦ ∈ F
t
, for all t ≥ 0.
Then τ is not in general a stopping time with respect to (F
t
)
t≥0
, but it is a
stopping time with respect to the ﬁltration (F
t
+)
t≥0
. In fact
¦τ ≤ t¦ =
∞
k=1
_
τ ≤ t +
1
k
_
∈ F
t
+.
Markov property 47
Exercise 4.7 Assume that the nonnegative random variable τ is discrete,
that is that τ(Ω) = (µ
k
)
k∈N
where µ
k
is an increasing sequence of positive
numbers. Show that τ is a stopping time if and only if ¦τ = µ
k
¦ ∈ F
µ
k
for
all k ∈ N. Show that in this case F
τ
is the σ–algebra
F
τ
: = ¦A ∈ F : A ∩ ¦τ = µ
k
¦ ∈ F
µ
k
for all k ∈ N¦.
Proposition 4.8 Let τ be a stopping time. Then there exists a decreasing
sequence (τ
n
) of discrete stopping times convergent pointwise to τ such that
F
τ
n
⊃ F
τ
for all n ∈ N.
Proof. Deﬁne for any n ∈ N and ω ∈ Ω
τ
n
(ω) =
k
2
n
if
k −1
2
n
≤ τ(ω) <
k
2
n
, k ∈ N. (4.2)
It is clear that the sequence (τ
n
) is decreasing. Moreover, τ
n
is a stopping
time. In fact, if t =
k
2
n
with k ∈ N we have
¦τ
n
= t¦ =
_
k −1
2
n
≤ τ <
k
2
n
_
∈ F
t
. (4.3)
Finally, let A ∈ F
τ
, that is
A ∩ ¦τ ≤ t¦ ∈ F
t
, ∀ t ≥ 0.
Then we have
A ∩
_
τ
n
=
k
2
n
_
= A ∩
_
k −1
2
n
≤ τ <
k
2
n
_
∈ F k
2
n
, ∀ k ∈ N,
so that A ∈ F
τ
n
.
We want to extend several properties concerning time t to general stop
ping times τ. We start by showing that W
τ
is F
τ
measurable.
Proposition 4.9 Let τ be a stopping time and set
W
τ
(ω) = W(τ(ω), ω), ω ∈ Ω.
Then W
τ
is F
τ
measurable.
Proof. Assume ﬁrst τ discrete,
τ(Ω) = ¦t
k
¦, 0 < t
1
< < t
k
<
48 Chapter 4
and set A
k
= ¦τ = t
k
¦, k ∈ N. Then we have
W
τ
(ω) = W(t
k
)(ω), ∀ω ∈ A
k
, k ∈ N.
Let I ∈ B(R). Then
¦W
τ
∈ I¦ ∩ ¦τ ≤ t¦ =
∞
k=1
[¦W
τ
∈ I¦ ∩ ¦τ ≤ t¦ ∩ A
k
]
=
∞
k=1
[¦W
t
k
∈ I¦ ∩ ¦τ ≤ t¦ ∩ A
k
]
=
∞
{k∈N: t
k
≤t}
[¦W
t
k
∈ I¦ ∩ ¦τ ≤ t¦ ∩ A
k
] ∈ F
t
.
So, the conclusion holds in this case.
Let now τ be arbitrary, let τ
n
be deﬁned by (4.2) and set
W
τ
n
(ω) = W(τ
n
(ω), ω), ω ∈ Ω.
Since W is continuous we have
lim
n→∞
W
τ
n
(ω) = W
τ
(ω), ω ∈ Ω.
Fix t ≥ 0. By the previous argument we have
¦W
τ
n
∈ I¦ ∩ ¦τ
n
≤ t¦ ∈ F
t
for all I ∈ B(R). (4.4)
Now the conclusion follows letting n → ∞.
Example 4.10 Let a ∈ R and set
(1)
τ
a
= inf¦t ≥ 0 : W(t) = a¦.
Then
¦τ
a
> t¦ =
s∈[0,t]
¦W(s) < a¦ =
s∈[0,t]∩Q
¦W(s) < a¦ ∈ F
t
.
So, τ
a
is a stopping time with respect to the ﬁltration (F
t
)
t≥0
.
Let now
τ = inf¦t ≥ 0 : W(t) > a¦.
Then we have
¦τ ≥ t¦ =
s∈[0,t]
¦W(s) ≤ a¦ =
s∈[0,t]∩Q
¦W(s) ≤ a¦ ∈ F
t
.
Consequently, by Remark 4.6, τ is a stopping time with respect to ﬁltration
¦F
t
+¦
t≥0
.
(1)
We use the convention that the inﬁmum of the empty set is +∞.
Markov property 49
4.3 The Brownian motion W(t + τ) −W(τ)
We recall that W(t +h) −W(t), t ≥ 0, is a Brownian motion for any h > 0.
We want now to show that the same holds when h is replaced by a stopping
time.
Proposition 4.11 Let τ be a stopping time. Then
C(t) := W(t + τ) −W(τ), t ≥ 0,
is a Brownian motion.
Proof. Let us ﬁrst prove that the law of C(t) is N
t
. For this it is enough to
show that for any α ∈ R we have
E
_
e
iαC(t)
_
= E
_
e
iα(W(t+τ)−W(τ))
_
= e
−
1
2
α
2
t
, α ∈ R. (4.5)
Assume ﬁrst that τ is discrete, τ(Ω) = (t
k
) and set
A
i
= ¦τ = t
i
¦ ∈ F
t
i
, ∀ i ∈ N.
Then we have
E
_
e
iα(W(t+τ)−W(τ))
_
=
∞
i=1
_
A
i
e
iα(W(t+t
i
)−W(t
i
))
dP =
∞
i=1
E
_
1l
A
i
e
iα(W(t+t
i
)−W(t
i
))
_
.
Since 1l
A
i
and W(t + t
i
) −W(t
i
) are independent, it follows that
E
_
e
iα(W(t+τ)−W(τ))
_
=
∞
i=1
P(A
i
)E
_
e
iα(W(t+t
i
)−W(t
i
))
_
= e
−
1
2
α
2
t
and so (4.5) is proved.
Let now τ be general and let (τ
n
) be the sequence of ﬁnite stoppping
times deﬁned by (4.2). We have just proved that
E
_
e
iα(W(t+τ
n
)−W(τ
n
))
_
= e
−
1
2
α
2
t
, α ∈ R.
Now (4.5) follows letting n tend to inﬁnity. By (4.5) it follows that C(t) is a
Gaussian random variable N
t
. Proceeding similarly one can prove that the
law of C(t) − C(s) with t > s > 0 is N
t−s
and that C(t) has independent
increments. Continuity of C(t) is obvious.
50 Chapter 4
4.4 Transition semigroup
We shall denote by B
b
(R) the set of all real, bounded and Borel functions
and by C
b
(R) the subspace of B
b
(R) of those functions which are uniformly
continuous and bounded on R.
Given ϕ ∈ B
b
(R) we want to study the evolution in time of ϕ(W(t) +x).
To this purpose, we deﬁne the transition semigroup
P
t
ϕ(x) = E[ϕ(W(t) + x)], t ≥ 0, x ∈ R, ϕ ∈ B
b
(R), (4.6)
Since the law of W(t) + x is N
x,t
we have
P
t
ϕ(x) = E[ϕ(W(t) + x)]
=
1
√
2πt
_
+∞
−∞
e
−
1
2t
(x−y)
2
ϕ(y)dy
=
_
+∞
−∞
g
t
(x −y)ϕ(y)dy,
(4.7)
where
g
t
(ξ) =
1
√
2πt
e
−
ξ
2
2t
, t > 0, ξ ∈ R. (4.8)
We deduce, by an explicit computation, that P
t
, t ≥ 0, is a semigroup of
linear operators in B
b
(R), that is P
0
= I and
P
t+s
= P
t
P
s
, ∀ t, s ≥ 0.
Notice that P
t
coincides with the heat semigroup in R. In fact one checks
easily that if ϕ ∈ C
b
(R) then the function u : [0, +∞) R → R, u(t, x) =
P
t
ϕ(x) is continuous, inﬁnitely diﬀerentiable and fulﬁlls
_
¸
_
¸
_
u
t
(t, x) =
1
2
u
xx
(t, x), ∀ t > 0, x ∈ R,
u(0, x) = ϕ(x), ∀ x ∈ R.
Remark 4.12 One can show that u(t, x) = P
t
ϕ(x), t ≥ 0, x ∈ R, is the
unique solution of the Dirichlet problem above.
There is a simple deterministic proof based on maximum principle and a
stochastic proof, which we will present later, based on Itˆo’s formula.
Exercise 4.13 Prove that for t > s ≥ 0,
P
t−s
ϕ(x) = E[ϕ(W(t) −W(s) + x)], ϕ ∈ B
b
(H), x ∈ R. (4.9)
Markov property 51
4.5 Markov property
In this section we shall use several properties of conditional expectation, they
are recalled in Appendix A.
We are here concerned with the stochastic process
X(t) = X(t, x) = W(t) + x, t ≥ 0,
where x ∈ R.
Proposition 4.14 For any t > s > 0 and any ϕ ∈ B
b
(H) we have
E[ϕ(X(t))[F
s
] = (P
t−s
ϕ)(X(s)). (4.10)
Equivalently
_
A
ϕ(X(t))dP =
_
A
(P
t−s
ϕ)(X(s))dP, ∀ A ∈ F
s
. (4.11)
Moreover X() is a Markov process.
Proof. Set
X(t) = W(t) + x = (W(s) + x) + (W(t) −W(s)) =: U + V.
Notice that U is F
s
measurable and V is independent of F
s
. By Proposition
B.6 it follows that
E[ϕ(X(t))[F
s
] = E[ϕ(U + V )[F
s
] = h(U),
where (recall Exercise 4.13)
h(u) = E[ϕ(u + V )] = E[ϕ(u + W(t) −W(s))] = P
t−s
ϕ(u).
So, (4.10) is proved.
To prove the last statement notice that by Proposition B.3 we have
E[ϕ(X(t))[X(s)] = E[E[ϕ(X(t))[F
s
][X(s)]
= E[P
t−s
ϕ(X(s))[X(s)]
= P
t−s
ϕ(X(s)) = E[ϕ(X(t))[F
s
].
Exercise 4.15 Let s > 0, η a F
s
measurable random variable and ϕ ∈
B
b
(R). Show that
E[ϕ(W(t) + η[F
s
] = (P
t−s
ϕ(η)).
52 Chapter 4
4.5.1 Strong Markov property
We now consider conditional expectation with respect to F
τ
where τ is a
stopping time.
Proposition 4.16 Let τ be a stopping time and let t ≥ τ and ϕ ∈ B
b
(H).Then
we have
E[ϕ(X(t))[F
τ
] = (P
t−τ
ϕ)(X(τ)). (4.12)
Equivalently
_
A
ϕ(X(t))dP =
_
A
(P
t−τ
ϕ)(X(τ))dP, ∀ A ∈ F
τ
. (4.13)
Proof. We set x = 0 for simplicity, so that X(t) = W(t). Assume ﬁrst that
τ is of the form
τ(Ω) = (t
k
)
k∈N
.
Let A ∈ F
τ
. Then we have
_
A
(P
t−τ
ϕ)(W(τ))dP =
∞
i=1
_
A∩{τ=t
i
}
(P
t−τ
ϕ)(W(τ))dP
=
∞
i=1
_
A∩{τ=t
i
}
(P
t−t
i
ϕ)(W(t
i
))dP.
Therefore, by (4.10) and taking into account that by the deﬁnition of F
τ
we
have
A ∩ ¦τ = t
i
¦ ∈ F
t
i
, i = 1, ..., n,
we can write,
_
A
(P
t−τ
ϕ)(W(τ))dP =
∞
i=1
_
A∩{τ=t
i
}
(P
t−t
i
ϕ)(W(t
i
))dP
=
∞
i=1
_
A∩{τ=t
i
}
E[ϕ(W(t))[F
t
i
]dP
=
∞
i=1
_
A∩{τ=t
i
}
ϕ(W(t))dP =
_
A
ϕ(W(t))dP.
Therefore, (4.13) is proved.
Markov property 53
Let now τ be an arbitrary stopping time and let (τ
n
) be deﬁned by (4.2).
Recall that (Proposition 4.8)
F
τ
⊂ F
τ
n
for all n ∈ N.
Let A ∈ F
τ
. Then by (4.13) it follows that
_
A
ϕ(W(t))dP =
_
A
(P
t−τ
n
ϕ)(W(τ
n
))dP for all A ∈ F
τ
.
Now the conclusion follows letting n → ∞.
Property (4.12) is called the strong Markov property of W.
4.6 Some consequences of the strong Markov
property
In this section we want to determine the laws of the following important
random variables.
• T
b
= inf¦t ≥ 0 : B(t) = b¦, b ∈ R.
• M(t) = max
s∈[0,t]
B(s), t ≥ 0.
• m(t) = min
s∈[0,t]
B(s), t ≥ 0.
Notice that
¦T
a
≤ t¦ = ¦M(t) ≥ a¦, t ≥ 0, a ≥ 0 (4.14)
and
¦T
a
≤ t¦ = ¦m(t) ≤ a¦, t ≥ 0, a ≤ 0. (4.15)
To ﬁnd the laws of T
a
with a ≥ 0 and M(t) the following lemma is useful.
Lemma 4.17 Let a ≥ 0 and t ≥ 0. Then we have
P(B(t) ≤ a, M(t) ≥ a) = P(B(t) ≥ a). (4.16)
Proof. We have, taking into account that
¦T
a
≤ t¦ = ¦M(t) ≥ a¦
54 Chapter 4
P(W(t) ≤ a, M(t) ≥ a) = P(W(t) ≤ a, T
a
≤ t)
=
_
{T
a
≤t}
1l
(−∞,a]
(W(t))dP
=
_
{T
a
≤t}
E[1l
(−∞,a]
(W(t))[F
T
a
]dP,
since ¦T
a
≤ t¦ ∈ F
T
a
. By the strong Markov property it follows that
P(W(t) ≤ a, M(t) ≥ a) =
_
{T
a
≤t}
E[1l
(−∞,a]
(W(t))[F
T
a
]dP
=
_
{T
a
≤t}
E[P
t−T
a
1l
(−∞,a]
(W(T
a
))]dP
=
_
{T
a
≤t}
E[P
t−T
a
1l
(−∞,a]
(a)]dP.
On the other hand, we have, as easily checked,
P
s
1l
(−∞,a]
(a) = P
s
1l
[a,+∞)
(a), ∀ s > 0, a > 0.
Therefore
P(W(t) ≤ a, M(t) ≥ a) =
_
{T
a
≤t}
E[P
t−T
a
1l
(−∞,a]
(a)]dP
=
_
{T
a
≤t}
E[P
t−T
a
1l
[a,+∞)
(a)]dP
=
_
{T
a
≤t}
E[1l
[a,+∞)
(W(t))[F
T
a
]dP
= P(W(t) ≥ a, M(t) ≥ a)
= P(W(t) ≥ a).
Proposition 4.18 (Reﬂection principle) For all a ≥ 0 we have
P(M(t) ≥ a) = 2P(W(t) ≥ a), (4.17)
Proof. Write
P(M(t) ≥ a) = P(M(t) ≥ a, W(t) ≤ a) +P(M(t) ≥ a, W(t) ≥ a).
Markov property 55
Now, by Lemma 4.17 we have P(M(t) ≥ a, W(t) ≤ a) = P(W(t) ≥ a).
Moreover, it is clear that P(M(t) ≥ a, W(t) ≥ a) = P(W(t) ≥ a) so, the
conclusion follows.
By Proposition 4.18 we can easily deduce the expressions of the laws of
M(t) and T
a
for all a ∈ R.
Corollary 4.19 (Law of M(t)) For all t ≥ 0 we have
(M(t)
#
P)(dξ) =
2
√
2πt
e
−
ξ
2
2t
1l
[0,+∞)
(ξ)dξ. (4.18)
Proof. We have in fact by Proposition 4.18 for any a ≥ 0
P(M(t) ≥ a) = 2P(W(t)[ ≥ a) =
2
√
2πt
_
+∞
a
e
−
ξ
2
2t
dξ
= P([W(t)[ ≥ a).
Remark 4.20 From Corollary 4.19 it follows that at ﬁxed time t the law
of M(t) coincides with that of [W(t)[, though random variables M(t) and
[W(t)[ are diﬀerent; in particular M(t) is increasing whereas [W(t)[ is not.
Obviously the laws of M() and [W()[ on C
0
([0, +∞)) are diﬀerent.
Corollary 4.21 (Law of T
a
) Let a ≥ 0 and t ≥ 0. Then we have
((T
a
)
#
P)(dt) =
a
√
2πt
3
e
−
a
2
2t
dt. (4.19)
Proof. By (4.14) and Proposition 4.18 we have
P(T
a
≤ t) = P(M(t) ≥ a) =
2
√
2πt
_
+∞
a
e
−
ξ
2
2t
dξ
=
2
√
2π
_
+∞
at
−1/2
e
−
η
2
2
dξ.
Therefore
d
dt
P(T
a
≤ t) =
a
√
2πt
3
e
−
a
2
2t
dt,
which implies the conclusion.
The following results can be proved similarly.
56 Chapter 4
Lemma 4.22 Let a ≤ 0 and t ≥ 0. Then we have
P(W(t) ≥ a, m(t) ≤ a) = P(W(t) ≤ a). (4.20)
Proposition 4.23 (Reﬂection principle) For all a ≤ 0 we have
P(m(t) ≤ a) = 2P(W(t) ≤ a). (4.21)
Corollary 4.24 (Law of m(t)) For all t ≥ 0 we have
(m(t)
#
P)(dξ) = −
2
√
2πt
e
−
ξ
2
2t
1
(−∞,a]
(ξ)dξ. (4.22)
Corollary 4.25 (Law of T
a
) Let a ∈ R and t ≥ 0. Then we have
((T
a
)
#
P)(dt) =
[a[
√
2πt
3
e
−
a
2
2t
dt. (4.23)
4.7 Application to partial diﬀerential equa
tions
For any x ≥ 0 we set in this section
τ
x
= inf¦t ≥ 0 : W(t) + x = 0¦ = T
−x
.
Moreover we consider the following processes which take values in [0, +∞).
(i) Y (t) = W(t) + x, ∀ t ∈ [0, τ
x
].
Y (t) is called the Brownian motion killed in 0.
(ii) U(t) = [W(t) + x[, x ≥ 0, t ≥ 0.
U(t) is called the Brownian motion reﬂected in 0
(iii) V (t) = W(t ∧ τ
x
) + x, t ≥ 0. V (t) is called the Brownian motion
absorbed in 0
Markov property 57
4.7.1 The Dirichlet problem in the halfline
We are here concerned with the process Y (t) = W(t) + x, ∀ t ∈ [0, τ
x
].
Deﬁne for any ϕ ∈ B
b
([0, +∞))
U
t
ϕ(x) := u(t, x) := E[ϕ(W(t) + x)1l
t≤τ
x
], t ≥ 0, x ∈ H. (4.24)
We are going to show that u(t, x) is the solution of the Dirichlet problem in
[0, +∞),
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
u
t
(t, x) =
1
2
u
xx
(t, x), x > 0, t > 0
u(t, 0) = 0, t > 0,
u(0, x) = ϕ(x), x ≥ 0.
(4.25)
Proposition 4.26 We have
u(t, x) =
_
+∞
0
[g
t
(x −y) −g
t
(x + y)]ϕ(y)dy, x ≥ 0, t ≥ 0, (4.26)
where g is deﬁned by (4.8).
Proof. We have
u(t, x) = E[ϕ(W(t) + x)1l
t≤τ
x
]
= P
t
ϕ(x) −E[ϕ(W(t) + x)1l
t>τ
x
],
where ϕ is extended to R by setting
ϕ(−x) = ϕ(x), x ≥ 0.
Write
E[ϕ(W(t) + x)1l
t>τ
x
] = E[E[1l
t>τ
x
ϕ(W(t) + x)[F
τ
x
]]
= E[1l
t>τ
x
E[ϕ(W(t) + x)[F
τ
x
]]
Now, using the strong Markov property we ﬁnd that,
E[ϕ(W(t) + x)1l
t>τ
x
] = E[1l
t>τ
x
(P
t−τ
x
ϕ)(0)] =: E[ψ(τ
x
)],
where
ψ(λ) = 1l
t>λ
1
_
2π(t −λ)
_
R
e
−
ξ
2
2(t−λ)
ϕ(ξ)dξ, λ > 0.
58 Chapter 4
Next, recalling the law of τ
x
(see (4.23)) it follows that
E[ϕ(W(t) + x)1l
t>τ
x
] =
_
t
0
__
R
g
t−s
(y)ϕ(y)dy
_
x
√
2πs
3
e
−
x
2
2s
ds
=
∂
∂x
_
t
0
__
R
g
t−s
(y)ϕ(y)dy
_
g
s
(x)ds
=
_
R
g
t
(x −y)ϕ(y)dy +
∂
∂x
_
R
G
x,y
ϕ(y)dy,
where
(2)
G
x,y
=
_
t
0
g
t−s
(y)g
s
(x)ds =
1
2
Erfc
_
[x[ +[y[
√
2t
_
.
Since, for x > 0,
∂
∂x
G
x,y
= −
1
√
2πt
e
−
(x+y)
2
2t
= −g
t
(x +[y[)
we get
u(t, x) =
_
R
g
t
(x −y)ϕ(y)dy −
_
R
g
t
(x +[y[)ϕ(y)dy,
and the conclusion follows.
It is easy to check, by a direct computation, that if ϕ ∈ C
b
([0, +∞)),
U
t
ϕ(x) = u(t, x) is the solution of the Dirichlet problem (4.25). Moreover
U
0
= I and U
t+s
= U(t)U(s) for all t, s ≥ 0.
4.7.2 The Neumann problem
We consider the process
U(t) = [W(t) + x[, x ≥ 0, t ≥ 0.
For any ϕ ∈ B
b
([0, +∞)) we set
Q
t
ϕ(x) = E[ϕ([W(t) + x[)] = (2πt)
−1/2
_
R
e
−
x−y
2
2t
ϕ([y[)dy.
Replacing in the last integral y with −y, we see that
Q
t
ϕ(x) =
_
+∞
0
[g
t
(x −y) + g
t
(x + y)]ϕ(y),
(2)
We recall that Erfc (a) =
2
√
π
_
+∞
a
e
−r
2
dr.
Markov property 59
where g
t
is deﬁned by (4.8).
Now it is easy to check that if ϕ ∈ C
b
([0, +∞)) then u(t, x) = Q
t
ϕ(x) is
continuous in [0, ∞) [0, ∞), inﬁnitely diﬀerentiable in (0, ∞) [0, ∞) and
solves the following Neumann problem
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
u
t
(t, x) =
1
2
u
xx
(t, x), x ≥ 0, t > 0,
u
x
(t, 0) = 0, t > 0,
u(0, x) = ϕ(x), x ≥ 0.
Moreover Q
0
= I and Q
t+s
= Q(t)Q(s) for all t, s ≥ 0.
4.7.3 The Ventzell problem
Let us consider the stochastic process,
V (t) = W(t ∧ τ
x
) + x, t ≥ 0,
where x ≥ 0.
Set
Z
t
ϕ(x) = E[ϕ(W(t ∧ τ
x
) + x)], ϕ ∈ B
b
([0, +∞)), x ≥ 0.
So,
Z
t
ϕ(x) =
_
Ω
ϕ(B(t ∧ τ
x
) + x)dP
=
_
{t<τ
x
}
ϕ(W(t) + x)dP +
_
{t≥τ
x
}
ϕ(0)dP,
since W(τ
x
) + x = 0. Therefore
Z
t
ϕ(x) = U
t
ϕ(x) + ϕ(0) P(T
−x
≤ t),
where U
t
is deﬁned by (4.24). So
Z
t
ϕ(x) =
_
+∞
0
[g
t
(x −y) −g
t
(x + y)]ϕ(y)dy +
ϕ(0)
√
2πt
_
x
−∞
e
−
y
2
2t
dy.
If ϕ ∈ C
b
([0, +∞)), setting u(t, x) = Z
t
ϕ(x) we see that u is the solution to
the Ventzell problem,
_
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
_
u
t
(t, x) =
1
2
u
xx
(t, x), x ≥ 0, t ≥ 0
u
xx
(t, 0) = 0, t ≥ 0,
u(0, x) = ϕ(x), x ≥ 0.
60 Chapter 4
Moreover Z
0
= I and Z
t+s
= Z(t)Z(s) for all t, s ≥ 0.
Chapter 5
The Itˆ o integral
In all this chapter B represents a Brownian motion in a probability space
(Ω, F, P).
Similarly as in Chapter 4, for any t > 0 we denote by C
t
the algebra of
all cylindrical sets
C
t
1
,··· ,t
n
;A
= ¦ω ∈ C
0
: (B(t
1
), ..., B(t
n
)) ∈ A¦
where 0 ≤ t
1
< ... < t
n
, t
n
≤ t and A ∈ B(R
n
).
Moreover, we denote by F
t
the σalgebra generated by C
t
and all Pnull
sets of Ω. We call F
t
, t ≥ 0 the natural ﬁltration of B augmented with the
null sets of P.
The family of σ–algebras (F
t
)
t≥0
is increasing; it is called the natural
ﬁltration of B.
We denote by (F
t
)
t≥0
the completion of the natural ﬁltration of B with
all Pnull sets of Ω.
We say that a stochastic process F(t), t ∈ [0, T], is adapted to the Brow
nian motion B if F(t) is F
t
measurable for any t ∈ [0, T].
5.1 Deﬁnition of Itˆo’s integral
5.1.1 Itˆ o’s integral for elementary processes
Deﬁnition 5.1 Let T > 0. An elementary process F(t), t ∈ [0, T], in
(Ω, F, P) is a stochastic process of the form
F =
n
i=1
F
i−1
1l
[t
i−1
,t
i
)
, (5.1)
61
62 The Itˆo integral
where n ∈ N, 0 = t
0
< t
1
< < t
n
= T and F
i
is F
t
i
measurable for any
i = 0, 1, ..., n −1.
For any elementary process F(t), t ∈ [0, T], we deﬁne the Itˆo integral
setting
I(F): =
_
T
0
F(s)dB(s) =
n
i=1
F
i−1
(B(t
i
) −B(t
i−1
)). (5.2)
Obviously any elementary process is adapted. This property is needed to
prove some basic identities (similar to those obtained for the Wiener integral)
which allow to extend the integral to more general processes.
Proposition 5.2 Assume that F ∈ E
2
B
(0, T). Then I(F) ∈ L
2
(Ω, F, P) and
we have
E
__
T
0
F(s)dB(s)
_
= 0 (5.3)
E
_
__
T
0
F(s)dB(s)
_
2
_
=
_
T
0
E([F(s)[
2
)ds. (5.4)
Proof. Let us prove (5.3). We have
E[I(F)] =
n
j=1
E[F
j−1
(B(t
j
) −B(t
j−1
))].
Since F
j−1
is F
j−1
measurable, it is independent of B(t
j
)−B(t
j−1
), by Lemma
4.3. Therefore we have
E[I(F)] =
n
j=1
E[F
j−1
]E[B(t
j
) −B(t
j−1
)] = 0
and (5.3) is proved.
Let us prove (5.4). We have
E[[I(F)[
2
] = E
_
n
j=1
[F
j−1
[
2
[B(t
j
) −B(t
j−1
)]
2
_
+2E
_
j<k
F
j−1
F
k−1
[B(t
j
) −B(t
j−1
)] [B(t
k
) −B(t
k−1
)]
_
.
Notice now that for j < k the random variable
F
j−1
F
k−1
[B(t
j
) −B(t
j−1
)],
Chapter 5 63
is F
k−1
–measurable and consequently is independent of B(t
k
) − B(t
k−1
).
Therefore, taking the expectation, we have
E[F
j−1
F
k−1
[B(t
j
) −B(t
j−1
)][B(t
k
) −B(t
k−1
)]]
= E[F
j−1
F
j−1
[B(t
j
) −B(t
j−1
)]] E[B(t
k
) −B(t
k−1
)] = 0.
It follows that
E[[I(F)[
2
] =
n
j=1
E[[F
j−1
[
2
](t
j
−t
j−1
),
as required.
Exercise 5.3 Let F, G ∈ E
2
B
(0, T). Prove that
E
__
T
0
F(s)dB(s)
_
T
0
G(s)dB(s)
_
=
_
T
0
E[F(s)G(s)]ds.
Hint: Use the identity
ab =
1
2
(a + b)
2
−
1
2
a
2
−
1
2
b
2
, a, b ∈ R.
5.1.2 General deﬁnition of Itˆ o’s integral
Let us denote by
Z
T
:= L
2
([0, T] Ω, B(0, T) F, dt P)
the Hilbert space of all (equivalence classes of) functions
F : [0, T] Ω, (t, ω) → F(t, ω),
which are measurable with respect to the product σalgebra, B(0, T) F
and such that
F
Z
T
:= E
_
T
0
[F(t, )[
2
dt < ∞.
The scalar product on Z is deﬁned by
¸F, F
1
) = E
_
T
0
F(t, )F
1
(t, )dt.
Obviously any elementary process F belongs to Z.
64 The Itˆo integral
In view of (5.4), the mapping
E
2
B
(0, T) ⊂ Z
T
→ L
2
(Ω, F
T
, P)F →
_
T
0
F(s)dB(s),
is an isometry. Therefore it can be uniquely extended to the closure E
2
B
(0, T)
of E
2
B
(0, T) in Z
T
.
Processes belonging to E
2
B
(0, T) are called predictable.
So, the Itˆ o integral can be uniquely deﬁned by extension for any pre
dictable square integrable process F(t), t ≥ 0 and the following properties
are fulﬁlled.
E
__
T
0
F(s)dB(s)
_
= 0 (5.5)
E
_
__
T
0
F(s)dB(s)
_
2
_
=
_
T
0
E([F(s)[
2
)ds. (5.6)
Moreover, from Exercise 5.3 it follows that if F and G are predictable square
integrable processes we have
E
__
T
0
F(s)G(s)dB(s)
_
=
_
T
0
E[F(s)G(s)]ds. (5.7)
We can deﬁne in an obvious way the Itˆo integral
_
b
a
F(s)dB(s) in any
interval [a, b] ⊂ [0, T]. We have
E
__
b
a
F(s)dB(s)
_
= 0,
and
E
_
__
b
a
F(s)dB(s)
_
2
_
=
_
b
a
(E[F(s)[
2
)ds.
Moreover, for any a, b, c ∈ [0, T] we have
_
c
a
F(s)dB(s) =
_
b
a
F(s)dB(s) +
_
c
b
F(s)dB(s).
Let us now present a characterization of predictable processes (that is of
space E
2
B
(0, T)). Note ﬁrst that an elementary process is a linear combination
of processes of the form
F1l
[a,b)
, with F F
a
measurable.
Chapter 5 65
In turn each F can be approximated by linear combinations of characteristic
functions of F
a
measurable sets. So, it is natural to approximate a general
predictable process by linear combinations of functions of the form
1l
A×[a,b)
, with A F
a
measurable.
We call A [a, b) a predictable rectangle. We denote by R the family of all
predictable rectangles and by P the σalgebra generated by R. P is called
the σalgebra of all predictable events.
Deﬁnition 5.4 A real predictable process in [0, T] is a real random variable
in the probability space
([0, T] Ω, P, dt P).
Proposition 5.5 The closure E
2
B
([0, T]) is precisely L
2
([0, T]Ω, P, dtP).
Proof. Denote by Λ
T
the closure of E
2
B
([0, T]) in L
2
([0, T] Ω, P, dt P).
Since any element of L
2
([0, T] Ω, P, dt P) can be approximated by a
monotonic sequence of simple functions, it is enough to show that 1l
A
∈ Λ
T
for any A ∈ P. For this we shall use the Dynkin Theorem, see Appendix A.
We ﬁrst note that R is a πsystem. Then we set
D = ¦A ∈ P : 1l
A
∈ Λ
T
¦.
We claim that D is a λsystem, i.e. that it fulﬁlls (A.1). Properties (B.1)
(i)(ii) are clear, let us show (A.1)(iii). Let (A
n
) ⊂ D be mutually disjoint
sets and set
φ
n
=
n
k=1
1l
A
k
.
Then, by the monotone convergence theorem, φ
n
→ φ = 1l
A
in L
2
([0, T]
Ω, P, dt P) where A =
∞
k=1
A
k
. So, A ∈ D and (A.1)(iii) is fulﬁlled. Now
the conclusion follows by Theorem A.1.
Exercise 5.6 Let F ∈ L
2
([0, T] Ω, P, dt P), [s, t] ⊂ [0, T] and let ϕ ∈
L
∞
(Ω, F
s
, P). Prove that
ϕ
_
t
s
F(r)dB(r) =
_
t
s
ϕ F(r)dB(r). (5.8)
Exercise 5.7 Let F ∈ L
2
([0, T] Ω, P, dt P) such that
_
T
0
F(s)dB(s) = 0.
Show that F = 0.
66 The Itˆo integral
5.2 Itˆ o integral for mean square continuous
processes
We shall denote by C
B
([0, T]; L
2
(Ω)) the space of all stochastic processes
which are mean square continuous and adapted. We recall that if F ∈
C
B
([0, T]; L
2
(Ω)) then F(t) is F
t
measurable for all t ∈ [0, T] and the map
ping
[0, T] → L
2
(Ω, F, P), t → F(t),
is continuous.
For any decomposition σ = ¦t
0
, t
1
, , t
n
¦ ∈ Σ(0, T) consider the ele
mentary process
F
σ
:=
n
j=1
F(t
j−1
)1l
[t
j−1
,t
j
)
and set
I
σ
(F) :=
_
T
0
F
σ
(s)dB(s) =
n
j=1
F(t
j−1
)(B(t
j
) −B(t
j−1
)).
Clearly F
σ
∈ E
2
B
(0, T) and, using the continuity of F one can check easily
that
lim
σ→0
F
σ
= F, in L
2
([0, T] Ω, P, dt P). (5.9)
Consequently we have
lim
σ→0
I
σ
(F) =
_
T
0
F(s)dB(s) in L
2
(Ω, F, P). (5.10)
Example 5.8 Let us prove that
_
T
0
B(t)dB(t) =
1
2
(B
2
(T) −T). (5.11)
Let σ = ¦t
0
, t
1
, ..., t
n
¦ ∈ Σ(0, T). Write
B(t
k−1
)(B(t
k
) −B(t
k−1
)) = B(t
k−1
)B(t
k
) −B
2
(t
k−1
))
= −
1
2
B
2
(t
k
) + B(t
k−1
)B(t
k
) −
1
2
B
2
(t
k−1
) +
1
2
B
2
(t
k
) −
1
2
B
2
(t
k−1
)
=
1
2
B
2
(t
k
) −
1
2
B
2
(t
k−1
) −
1
2
(B(t
k
) −B(t
k−1
))
2
.
Chapter 5 67
Then we have
I
σ
(B) =
1
2
B
2
(T) −
1
2
n
k=1
(B(t
k
) −B(t
k−1
))
2
.
Recalling that the quadratic variation of B is T (Theorem 3.19), we deduce
that
_
T
0
B(t)dB(t) = lim
σ→0
I
σ
(B) =
1
2
(B
2
(T) −T).
Exercise 5.9 Prove that
lim
σ→0
n
k=1
B(t
k
)(B(t
k
) −B(t
k−1
)) =
1
2
(B
2
(T) + T), in L
2
(Ω, F, P),
and
lim
σ→0
n
k=1
B
_
t
k
+ t
k−1
2
_
(B(t
k
) −B(t
k−1
)) =
1
2
B
2
(T), in L
2
(Ω, F, P).
Therefore the deﬁnition of the Itˆo integral depends on the particular form of
the integral sums.
5.3 The Itˆ o integral as a stochastic process
Let F ∈ L
2
([0, T] Ω, P, dt P and set
X(t) =
_
t
0
F(s)dB(s), t ∈ [0, T].
We ﬁrst notice that X(t), t ≥ 0, is not a process with independent increments
in general (unless f is deterministic); take for instance
X(t) =
_
t
0
B(s)dB(s) =
1
2
(B
2
(t) −t), t ≥ 0.
However, X(t), t ≥ 0, has orthogonal increments (in the sense of L
2
(Ω, F, P))
as the following result shows.
Proposition 5.10 Let 0 ≤ t
1
≤ t
2
≤ t
3
≤ t
4
≤ T. Then we have
E[(X(t
2
) −X(t
1
))(X(t
4
) −X(t
3
))] = 0
68 The Itˆo integral
Proof. We have in fact, taking into account (5.7)
E[(X(t
2
) −X(t
1
))(X(t
4
) −X(t
3
))]
= E
__
t
2
t
1
F(s)dB(s)
_
t
4
t
3
F(s)dB(s)
_
= E
__
T
0
1l
[t
1
,t
2
]
F(s)dB(s)
_
T
0
1l
[t
3
,t
4
]
F(s)dB(s)
_
=
_
T
0
1l
[t
1
,t
2
]
1l
[t
3
,t
4
]
E(F
2
(s))ds = 0.
We are going to show that X(t), t ≥ 0, is mean square continuous, then
that it is a continuous process.
Proposition 5.11 Let F ∈ L
2
([0, T]Ω, P, dtP). Then X ∈ C
B
([0, T]; L
2
(Ω)).
Proof. We know that for any t ∈ [0, T], X(t) ∈ L
2
(Ω, F
t
, P). Moreover, for
any t, t
0
∈ [0, T] we have
E([X(t) −X(t
0
)[
2
) =
¸
¸
¸
¸
_
t
t
0
E([F(r)[
2
)dr
¸
¸
¸
¸
,
so that
lim
t→t
0
E([X(t) −X(t
0
)[
2
) = 0.
The conclusion follows.
We show now that X(t), t ≥ 0, is a continuous process. For this we ﬁrst
prove that it is a martingale with respect to the ﬁltration (F
t
) (see Appendix
C).
Proposition 5.12 X(t), t ∈ [0, T], is a F
t
–martingale
Proof. Let t > s. Since
X(t) −X(s) =
_
t
s
F(r)dB(r),
we have
E[X(t)[F
s
] = X(s) +E
__
t
s
F(r)dB(r)[F
s
_
.
Chapter 5 69
So, it remains to prove that
E
__
t
s
F(r)dB(r)[F
s
_
= 0. (5.12)
Notice that this is not obvious since
_
t
s
F(r)dB(r) is not independent of F
s
in general
(1)
. It is enough to prove (5.12) when F is an elementary process,
F =
n
i=1
F
i−1
1l
[t
i−1
,t
i
)
,
where s = t
1
, , t
n
= t and F
i−1
∈ L
2
(Ω, F, P). In this case, taking into
account that F
s
⊂ F
i−1
, we write
E
__
t
s
F(r)dB(r)[F
s
_
=
n
i=1
E[F
i−1
(B(t
i
) −B(t
i−1
))[F
s
]
=
n
i=1
E¦E[F
i−1
(B(t
i
) −B(t
i−1
))[F
i−1
][F
s
¦ = 0,
since F
i−1
is F
i−1
–measurable and B(t
i
) − B(t
i−1
) is independent of F
i−1
.
So, (5.12) is proved and the conclusion follows.
We are now ready to prove the continuity of X.
Theorem 5.13 Let F ∈ L
2
([0, T] Ω, P, dt P) and let
X(t) =
_
t
0
F(s)dB(s), t ∈ [0, T].
Then X has a continuous version and
E
_
sup
t∈[0,T]
[X(t)[
2
_
≤ 4
_
T
0
E[F(s)[
2
ds. (5.13)
Proof. Let (F
n
) ⊂ E
2
B
(0, T) such that
F
n
→ F in L
2
([0, T] Ω, P, dt P)
and set
X
n
(t) =
_
t
0
F
n
(s)dB(s), n ∈ N, t ∈ [0, T].
(1)
because F(r) contains in general the “story” of the Brownian motion from 0 to r.
70 The Itˆo integral
Since B(t) is continuous it is clear that X
n
(t) is continuous for all n ∈
N. Taking into account Proposition 5.12 we see that X(t), t ∈ [0, T], is
a continuous F
t
–martingale. Then by Corollary C.6 it follows that for any
n, m ∈ N
E
_
sup
t∈[0,T]
[X
n
(t) −X
m
(t)[
2
_
≤ 4E([X
n
(T) −X
m
(T)[
2
)
= 4E
__
T
0
[F
n
(s) −F
m
(s)[
2
ds
_
.
Consequently (X
n
)(ω) is Cauchy in C([0, T]) for almost all ω and its limit,
which coincides with X(ω) is continuous.
5.4 Itˆ o integral with stopping times
5.4.1 Stopping times
We proceed here as in Section 4.2.
A nonnegative extended random variable τ in (Ω, F, P) is called a stopping
time with respect to the ﬁltration (F
t
)
t≥0
if
¦τ ≤ t¦ ∈ F
t
for all t ≥ 0.
To any stopping time τ we associate the σalgebra
F
τ
: = ¦A ∈ F : A ∩ ¦τ ≤ t¦ ∈ F
t
for all t ≥ 0¦.
The proofs of the two following propositions are completely similar to that
of Proposition 4.8 and 4.8. So, they will be omitted.
Proposition 5.14 Let τ be a stopping time. Then there exists a decreasing
sequence (τ
n
) of discrete stopping times convergent pointwise to τ such that
F
τ
n
⊃ F
τ
for all n ∈ N.
Proposition 5.15 Let τ be a stopping time and set
W(τ)(ω) = W(τ(ω))(ω), ω ∈ Ω.
Then W(τ) is F
τ
measurable and W(t + τ) − W(τ), t ≥ 0 is a Brownian
motion in (Ω, F, P).
Chapter 5 71
5.4.2 Itˆ o’s integral with stopping times
Let F ∈ L
2
([0, T] Ω, P, λ P) and set
X(t) =
_
t
0
F(s)dB(s), t ∈ [0, T].
Let moreover τ ≤ T be a stopping time. Deﬁne
_
τ
0
F(s)dB(s): = X(τ),
where
X(τ, ω) = X(τ(ω), ω), ω ∈ Ω.
Arguing as in Proposition 5.15 and using the fact that X(t), t ∈ [0, T], has
a continuous version, one can see that X(τ) is F
τ
–measurable.
The following result reduces a Itˆ o’s integral with a stopping time to a
usual one between 0 to T.
Proposition 5.16 Let F ∈ L
2
([0, T] Ω, P, dt P) and let τ ≤ T be a
stopping time. Then we have
_
τ
0
F(s)dB(s) =
_
T
0
1l
{s<τ}
F(s)dB(s). (5.14)
Proof. It is enough to prove the result when τ is of the form,
τ(Ω) = (t
1
, t
2
, ..., t
n
),
with 0 < t
1
< t
2
< < t
n
≤ T.
Set
A
i
:= ¦τ = t
i
¦, i = 1, ..., n.
T
¯
hen A
i
∈ F
t
i
, i = 1, ..., n.
Consider now the stochastic process
h(s) = 1l
{s≤τ}
, s ∈ [0, T].
We have
h(s) = 1, s ∈ [0, t
1
).
If s ∈ [t
1
, t
2
) we have
h(s)(ω) = 1 if ω ∈ A
2
∪ ∪ A
n
,
72 The Itˆo integral
so that
h(s) = 1l
A
2
∪···∪A
n
= 1l
A
c
1
.
Similarly, if s ∈ [t
k−1
, t
k
) with k ≤ n we have
h(s) = 1l
(A
k
∪...∪A
n
)
c.
Then h is predictable and
_
T
0
1l
{t<τ}
F(s)dB(s) =
_
t
1
0
F(s)dB(s) + 1l
(A
1
)
c
_
t
2
t
1
F(s)dB(s)
+ + 1l
(A
1
∪A
2
∪···∪A
n−1
)
c
_
t
n
t
n−1
F(s)dB(s)
= X(t
1
) + 1l
(A
1
)
c(X(t
2
) −X(t
1
))
+ + 1l
(A
1
∪A
2
∪···∪A
n−1
)
c(X(t
n
) −X(t
n−1
) = X(τ).
5.5 Multidimensional Itˆ o integrals
Let m ∈ N be ﬁxed and consider a standard mdimensional Brownian motion
B(t) = (B
1
(t), ..., B
m
(t)), t ≥ 0
in the probability space (Ω, F, P). Let (F
t
)
t∈[0,T]
be the natural ﬁltration of
B (augmented with all Pnull sets of Ω) .
We shall deﬁne the Itˆ o integral for predictable processes with values
in L(R
m
, R
d
) (that is such that any matrix element belongs to L
2
([0, T]
Ω, P, dtP)). We shall denote this space by L
2
([0, T]Ω, P, dtP; L(R
m
, R
d
))).
First we need a lemma whose simple proof is left to the reader.
Lemma 5.17 Let f, g ∈ L
2
([0, T] Ω, P, dt P). Then we have
E
__
T
0
f(s)dB
i
(s)
_
T
0
g(s)dB
j
(s)
_
= δ
i,j
_
T
0
E[f(s)g(s)]ds, i, j = 1, ..., m.
(5.15)
Let now F ∈ L
2
([0, T] Ω, P, dt P; L(R
m
, R
d
)). We deﬁne the Itˆo
integral of F as the ddimensional process
__
T
0
F(t)dB(t)
_
i
=
m
j=1
_
T
0
F
i,j
(t)dB
j
(t), i = 1, ..., d.
Chapter 5 73
Proposition 5.18 Let F ∈ L
2
([0, T] Ω, P, dt P; L(R
m
, R
d
)). Then we
have
E
¸
¸
¸
¸
_
T
0
F(t)dB(t)
¸
¸
¸
¸
2
=
_
T
0
E[Tr (F(t)F
∗
(t))]dt, (5.16)
where Tr denotes the trace.
Proof. Set I(F) =
_
T
0
F(t)dB(t). Then we have
(I(F))
i
=
m
j=1
_
T
0
F
i,j
(t)dB
j
(t), i = 1, ..., d.
It follows that
E[I(F)[
2
=
d
i=1
E
_
m
j=1
_
T
0
F
i,j
(t)dB
j
(t)
_
2
and, taking into account (5.15),
E[I(F)[
2
=
d
i=1
m
j=1
_
T
0
E[F
i,j
(t)
2
]dt,
which yields (5.16).
Remark 5.19 Assume that d = 1 so that L(R
d
; R
m
) is isomorphic to R
m
and F becomes a vector F = (F
1
, , F
m
).
In this case we shall write the Itˆo integral of F as
_
T
0
¸F(s), dB(s))
and formula (5.16) reduces to
E
¸
¸
¸
¸
_
T
0
¸F(t), dB(t))
¸
¸
¸
¸
2
=
_
T
0
E[F(t)[
2
dt. (5.17)
74 The Itˆo integral
Chapter 6
The Itˆ o formula
6.1 Introduction
Let (Ω, F, P) be a probability space, B a real Brownian motion, (F
t
)
t≥0
the
natural ﬁltration of B augmented with the null sets of P and P the σalgebra
of all predictable events (also augmented with the null sets of P).
We are given two stochastic processes b, σ ∈ L
2
([0, T] Ω, P, dt P) and
consider the stochastic process
X(t) = x +
_
t
0
b(s)ds +
_
t
0
σ(s)dB(s), t ≥ 0, (6.1)
where x ∈ R. X is adapted, continuous and continuous in mean square.
We set
dX(t) = b(t)dt + σ(t)dB(t)
and call dX(t) the Itˆo diﬀerential of X.
Given a regular real function ϕ, we are going to give a meaning to the
Itˆo’s diﬀerential ϕ
(X(t)).
We need some notations. For any k ∈ N we denote by C
k
b
(R) the linear
space of all real mappings which are uniformly continuous and bounded to
gether with their derivatives of order less or equal to k. If ϕ ∈ C
k
b
(R) we
set
ϕ
0
= sup
x∈R
[ϕ(x)[,
and
ϕ
k
= ϕ
0
+
k
j=1
sup
x∈R
[D
j
ϕ(x)[.
75
76 Chapter 6
We shall prove the following Itˆo’s formula,
ϕ(X(t)) = ϕ(x) +
_
t
0
ϕ
(X(s))σ(s)dB(s)
+
_
t
0
_
1
2
σ
2
(s)ϕ
(X(s)) + b(s)ϕ
(X(s))
_
ds, t ≥ 0.
(6.2)
We shall write (6.2) in the diﬀerential form, setting
ϕ
(X(t)) = ϕ
(X(t))σ(t)dB(t),
+
_
1
2
σ
2
(t)ϕ
(X(t)) + b(t)ϕ
(X(t))
_
dt, t ≥ 0,
(6.3)
or, also as
ϕ
(X(t)) = ϕ
(X(t))dX(t) +
1
2
σ
2
(t)ϕ
(X(t))dt, t ≥ 0. (6.4)
Remark 6.1 One can deduce formally Itˆo’s formula by proceeding as fol
lows. Write dX = b(t)dt + σ(t)dB and
dϕ(X) = ϕ(X + dX) −ϕ(X) = ϕ
(X)dX +
1
2
ϕ
(X)(dX)
2
= ϕ
(X)dX +
1
2
ϕ
(X)b
2
(t)(dt)
2
+ 2b(t)σ(t)dt dB + σ
2
(t)(dB)
2
.
Put (dB)
2
= dt and neglet the terms of order greater than dt, that is terms
with (dt)
2
and dt dB(t).
Writing (dB)
2
= dt is justiﬁed by Lemma 6.2 below.
Tthe following result on quadratic sums of a process is a generalization of
Theorem 3.19.
Lemma 6.2 Let F ∈ C
B
([0, T]; L
2
(Ω, F, P)) and let η = ¦0 = t
0
< t
1
<
< t
n
= T¦ ∈ Σ(0, T). Then we have
lim
η→0
n
k=1
F(t
k−1
)(B(t
k
) −B(t
k−1
))
2
=
_
T
0
F(s)ds in L
2
(Ω, F, P) (6.5)
Proof. Set
J
η
:=
n
k=1
F(t
k−1
)(B(t
k
) −B(t
k−1
))
2
.
The Itˆo formula 77
It is enough to prove that
lim
η→0
E
_
_
_
J
η
−
n
k=1
F(t
k−1
)(t
k
−t
k−1
)
_
2
_
_
= 0, (6.6)
since, obviously
lim
η→0
n
k=1
F(t
k−1
)(t
k
−t
k−1
) =
_
T
0
F(s)ds in L
2
(Ω, F, P).
To prove (6.6) write
E
_
_
_
J
η
−
n
k=1
F(t
k−1
)(t
k
−t
k−1
)
_
2
_
_
= E
_
_
_
n
k=1
F(t
k−1
)
_
[B(t
k
) −B(t
k−1
)[
2
−(t
k
−t
k−1
)
_
_
2
_
_
=
n
k=1
E
_
[F(t
k−1
)[
2
_
[B(t
k
) −B(t
k−1
)[
2
−(t
k
−t
k−1
)
¸
2
_
+2
n
j<k=1
E
_
F(t
j−1
)[[B(t
j
) −B(t
j−1
)[
2
−(t
j
−t
j−1
)]
F(t
k−1
)[[B(t
k
) −B(t
k−1
)[
2
−(t
k
−t
k−1
)]
_
Since the Brownian motion has independent increments, the last sum van
ishes, so that
E
_
_
_
J
η
−
n
k=1
F(t
k−1
)(t
k
−t
k−1
)
_
2
_
_
=
n
k=1
E
_
[F(t
k−1
)[
2
_
[B(t
k
) −B(t
k−1
)[
2
−(t
k
−t
k−1
)
¸
2
_
=
n
k=1
E[F(t
k−1
)[
2
E
_
_
[B(t
k
) −B(t
k−1
)[
2
−(t
k
−t
k−1
)
¸
2
_
,
(6.7)
78 Chapter 6
since F(t
k−1
) and B(t
k
) −B(t
k−1
) are independent.
Now, taking into account that
E[[B(t
k
) −B(t
k−1
)[
2
] = (t
k
−t
k−1
),
E[[B(t
k
) −B(t
k−1
)[
4
] = 3(t
k
−t
k−1
)
2
,
we have
E
_
_
_
J
η
−
n
k=1
F(t
k−1
)(t
k
−t
k−1
)
_
2
_
_
= 2
n
k=1
E[[F(t
k−1
)[
2
](t
k
−t
k−1
)
2
≤ 2[η[
n
k=1
E[[F(t
k−1
)[
2
(t
k
−t
k−1
)] → 0,
as [η[ → 0. The conclusion follows.
Now we are in position to prove Itˆ o’s formula. First we assume that b
and σ are elementary processes,
b =
p
i=1
b
i−1
1l
[λ
i−1
,λ
i
)
, σ =
p
i=1
σ
i−1
1l
[λ
i−1
,λ
i
)
, (6.8)
where p ∈ N, 0 = λ
0
< λ
1
< < λ
p
and b
i
, σ
i
are F
t
i
measurable for any
i = 0, 1, ..., p −1.
Lemma 6.3 Let ϕ ∈ C
2
b
(R), x ∈ R, b and σ given by (6.8) and X by (6.1).
Then identity (6.2) holds.
Proof. Since C
3
b
(R) is dense in C
2
b
(R) it is enough to show (6.2) when
ϕ ∈ C
3
b
(R). We start by proving (6.2) in [0, t] with t ≤ λ
1
. In this case we
have
b(t) = b
0
, σ(t) = σ
0
, t ∈ [0, λ
1
]
and
X(t) = b
0
t + σ
0
B(t), t ∈ [0, λ
1
].
Let η = ¦t
0
= 0 < t
1
< < t
N
= t¦. Then we obviously have
ϕ(X(t)) −ϕ(x) =
N
k=1
[ϕ(X(t
k
)) −ϕ(X(t
k−1
))].
The Itˆo formula 79
On the other hand, using Taylor’s formula we can write
ϕ(X(t)) −ϕ(x) =
N
k=1
ϕ
(X(t
k−1
))(X(t
k
) −X(t
k−1
))
+
1
2
N
k=1
ϕ
(X(t
k−1
))(X(t
k
) −X(t
k−1
))
2
+ R
η
=: I
1
+ I
2
+ I
3
. (6.9)
Concerning I
1
we have
I
1
=
N
k=1
ϕ
(X(t
k−1
))(b
0
(t
k
−t
k−1
) + σ
0
(B(t
k
) −B(t
k−1
)).
So,
lim
η→0
I
1
=
_
t
0
ϕ
(X(s))b(s)ds +
_
t
0
ϕ
(X(s))σ(s)dB(s) in L
2
(Ω, F, P).
(6.10)
Concerning I
2
we write
2I
2
=
N
k=1
ϕ
(X(t
k−1
))b
2
0
(t
k
−t
k−1
)
2
+ 2
N
k=1
ϕ
(X(t
k−1
))b
0
σ
0
(t
k
−
k−1
)(B(t
k
) −B(t
k−1
))
+
N
k=1
ϕ
(X(t
k−1
))σ
2
0
(B(t
k
) −B(t
k−1
))
2
=: I
2,1
+ I
2,2
+ I
2,3
. (6.11)
It is easy to check that
lim
η→0
I
2,1
= lim
η→0
I
2,2
= 0 in L
1
(Ω, F, P) (6.12)
In fact
[I
2,1
[ ≤
1
2
ϕ
2
[b
0
[
2
N
k=1
(t
k
−t
k−1
)
2
→ 0 as [η[ → 0
80 Chapter 6
and
(1)
E[I
2,2
[ ≤ ϕ
2
[b
0
[ [σ
0
[
N
k=1
(t
k
−t
k−1
)E[B(t
k
) −B(t
k−1
)[
≤ ϕ
2
[b
0
[ [σ
0
[
N
k=1
(t
k
−t
k−1
)
3/2
→ 0 as [η[ → 0.
Moreover, by Lemma 6.2 it follows that
lim
η→0
2I
2,3
=
_
t
0
ϕ
(X(s))σ
2
(s)ds in L
2
(Ω, F, P). (6.13)
So, the conclusion will follow provided
lim
η→0
E[R
η
[ = 0, (6.14)
Let us prove (6.14). We have
R
η
=
N
k=1
_
1
0
(1 −ξ)[ϕ
(ξ
k
) −ϕ
(X(t
k−1
))](X(t
k
) −X(t
k−1
))
2
dξ,
where
ξ
k
= (1 −ξ)X(t
k−1
) + ξX(t
k
).
Since ϕ ∈ C
3
b
(R) we have by the mean value theorem,
[ϕ
(ξ
k
) −ϕ
(X(t
k−1
))[ ≤ ϕ
0
(1 −ξ)[X(t
k
) −X(t
k−1
)[,
so that, we deduce setting 1 −ξ ≤ 1,
[R
η
[ ≤ ϕ
3
N
k=1
[X(t
k
) −X(t
k−1
)[
3
.
Consequently
[R
η
[ ≤ 3ϕ
3
[b
0
[
3
N
k=1
[t
k
−t
k−1
[
3
+ 3ϕ
3
[σ
0
[
3
N
k=1
[B(t
k
) −B(t
k−1
)[
3
(1)
since E[B(t)[ ≤ [E[B
2
(t)[]
1/2
= t
1/2
.
The Itˆo formula 81
and so
(2)
,
E([R
η
[) ≤ 3ϕ
3
[b
0
[
3
N
k=1
[t
k
−t
k−1
[
3
+ 3ϕ
3
[σ
0
[
3
√
15
N
k=1
[t
k
−t
k−1
[
3/2
→ 0,
as [η[ → 0. The proof is complete when t ≤ λ
1
. The general case can be
treated in the same way taking into account that b
k−1
and σ
k−1
are indepen
dent of B(t
k
) −B(t
k−1
).
We ﬁnally prove
Theorem 6.4 Let x ∈ R, b, σ ∈ L
2
([0, T] Ω, P, dt P) and ϕ ∈ C
2
b
(R).
Then identity (6.2) holds for all t ∈ [0, T].
Proof. Let (b
j
) and (σ
j
) be sequences of elementary processes such that
lim
j→∞
b
j
= b, lim
j→∞
σ
j
= σ in L
2
([0, T] Ω, P, dt P).
Set, for any j ∈ N,
X
j
(t) = x +
_
t
0
b
j
(s)ds +
_
t
0
σ
j
(s)dB(s), s ∈ [0, T]. (6.15)
Then we have (see (5.10))
lim
j→∞
X
j
= X in C
B
([0, T]; L
2
(Ω)).
Moreover by (6.2) we have
ϕ(X
j
(t)) = ϕ(x) +
_
t
0
ϕ
(X
j
(s))σ
j
(s)dB(s),
+
_
t
0
_
1
2
σ
j
(s)ϕ
(X
j
(s)) + b
j
(s)ϕ
(X
j
(s))
_
ds.
(6.16)
Now the conclusion follows by the dominated convergence theorem letting
j → ∞.
Taking expectation in the Itˆo formula we ﬁnd a useful identity which
allows to estimate the expectation of ϕ(X(t)).
(2)
Since E[B(t)[
3
) ≤ [E(B(t)
6
)]
1/2
=
√
15.
82 Chapter 6
Proposition 6.5 Assume that x ∈ R, b, σ ∈ L
2
([0, T] Ω, P, dt P) and
ϕ ∈ C
2
b
(R). Let
X(t) = x +
_
t
0
b(s)ds +
_
t
0
σ(s)dB(s), t ∈ [0, T].
Then
E[ϕ(X(t))] = ϕ(x) +
1
2
E
_
t
0
[ϕ
(X(s))σ
2
(s) + 2ϕ
(X(s))b(s)]ds. (6.17)
6.1.1 The Itˆ o formula for unbounded functions
We want now to show that formula (6.17) also holds without the assumption
that ϕ is bounded, provided the integrand in the right hand side is summable.
Proposition 6.6 Assume that x ∈ R, b, σ ∈ L
2
([0, T] Ω, P, dt P) and
ϕ ∈ C
2
(R). Set
X(t) = x +
_
t
0
b(s)ds +
_
t
0
σ(s)dB(s), t ∈ [0, T]. (6.18)
and assume in addition that
E
_
t
0
[ϕ
(X(s))σ
2
(s) + 2ϕ
(X(s))b(s)[ds < +∞. (6.19)
Then E[ϕ(X(t))] < +∞ and (6.17) holds.
Example 6.7 Take ϕ(x) = x
2
. Then condition (6.19) becomes
E
_
t
0
[σ
2
(s) + 2X(s)b(s)[ds < +∞
which is clearly fulﬁlled. Then
E([X(t)[
2
) = [x[
2
+E
_
t
0
(σ
2
(s) + 2X(s)b(s))ds.
Proof of Proposition 6.6. For any R > 0 consider a function ϕ
R
∈
C
2
b
(R) such that
ϕ
R
(x) =
_
ϕ(x) if [x[ ≤ R,
0 if [x[ ≥ R + 1.
The Itˆo formula 83
Then, applying Itˆo’s formula (6.2) to ϕ
R
(X(t)), yields for any R > 0
ϕ
R
(X(t)) −ϕ(x) =
1
2
_
t
0
[ϕ
R
(X(s))σ
2
(s) + 2ϕ
R
(X(s)b(s)]ds
+
_
t
0
ϕ
R
(X(s)))σ(s)dB(s).
(6.20)
Let now τ
R
be the stopping time
τ
R
=
_
¸
¸
_
¸
¸
_
inf¦t ∈ [0, T] : [X(t)[ ≥ R¦ if sup
t∈[0,T]
[X(t)[ ≥ R,
T if sup
t∈[0,T]
[X(t)[ < R.
It is clear that τ
R
is increasing and bounded by T. We know that X(, ω) is
continuous for almost all ω ∈ Ω. For such a ω, X(, ω) attains the maximum,
say M(ω). Then we have τ
R
(ω) = T for all R > M(ω). So,
lim
R→∞
τ
R
= T P–a.s.. (6.21)
Now, in view of Proposition 5.16 we can write
ϕ(X(t ∧ τ
R
)) −ϕ(x) =
1
2
_
t
0
1l
s<(t∧τ
R
)
[ϕ
(X(s))σ
2
(s) + 2ϕ
(X(s)b(s)]ds
+
_
t
0
1l
s<(t∧τ
R
)
ϕ
(X(s)))σ(s)dB(s).
(6.22)
Taking expectation we obtain
E[ϕ(X(t ∧ τ
R
))] −ϕ(x)
=
1
2
E
_
t
0
1l
s<(t∧τ
R
)
[ϕ
(X(s))σ
2
(s) + 2ϕ
(X(s)b(s)]ds.
(6.23)
Now, by the assumption (6.19), (6.21) and the dominated convergence theo
rem, we can let R → ∞ obtaining the conclusion.
As an application of Proposition 6.6 let us estimate E
_
_
T
0
F(s)dB(s)
_
2m
where F is predictable and m ∈ N, m > 1.
84 Chapter 6
Proposition 6.8 Assume that F ∈ L
2m
([0, T] Ω; P, dt P), m ∈ N, and
set
X(t) =
_
t
0
F(s)dB(s), t ∈ [0, T].
Then X ∈ L
2m
([0, T] Ω; P, dt P) and we have
E[[X(T)[
2m
] ≤ [m(2m−1)]
m
T
m−1
_
T
0
E
_
[F(t)[
2m
¸
dt. (6.24)
Proof. It is enough to prove (6.24) when F is bounded (because L
∞
([0, T]
Ω; P, dt P) is dense in L
2m
([0, T] Ω; P, dt P)).
We start from the case m = 2, setting ϕ(x) = x
4
. Then (6.19) holds so
that, by Proposition 6.6 we have
E[[X(t)[
4
] = 6E
__
t
0
[X(s)[
2
[F(s)[
2
ds
_
.
By H¨older’s inequality it follows that
E[[X(t)[
4
] ≤ 6
_
E
_
t
0
[X(s)[
4
ds
_
1/2
_
E
_
t
0
[F(s)[
4
ds
_
1/2
. (6.25)
Integrating between 0 and T, yields
_
T
0
E[X(t)[
4
dt ≤ 6T
_
E
_
T
0
[X(t)[
4
dt
_
1/2
_
E
_
T
0
[F(t)[
4
dt
_
1/2
. (6.26)
From which
_
T
0
E[X(t)[
4
dt ≤ 36T
2
_
T
0
E[F(t)[
4
dt.
Substituting this in (6.25) yields
E[[X(t)[
4
] ≤ 36TE
_
T
0
[F(t)[
4
dt.
So, (6.24) is proved for m = 2. We can now easily iterate the previous
argument taking successively m = 3, 4 and so on.
6.2 Itˆ o’ formula for a vector valued process
Let d, m ∈ N. Assume that x ∈ R
d
, b ∈ L
2
([0, T] Ω; P, dt P; R
d
) and
σ ∈ L
2
([0, T] Ω; P, dt P; L(R
m
; R
d
)). Set
X(t) = x +
_
t
0
b(s)ds +
_
t
0
σ(s)dW(s), t ∈ [0, T]
The Itˆo formula 85
We are going to prove the following Itˆo’s formula,
ϕ(X(t)) = ϕ(x) +
_
t
0
¸Dϕ(X(s)), σ(s)dB(s)),
+
_
t
0
_
1
2
Tr[(σσ
∗
)(s)D
2
ϕ(X(s))] +¸b(s), Dϕ(X(s)))
_
ds,
(6.27)
for all t ∈ [0, T]. We shall write (6.27) in the diﬀerential form
ϕ
(X(t)) = ¸Dϕ(X(t)), σ(t)dB(t))
+
_
1
2
Tr[(σσ
∗
)(t)D
2
ϕ(X(t))] +¸b(t), Dϕ(X(t)))
_
dt, t ≥ 0,
(6.28)
The proof is similar to that of the onedimensional case seen before. So, we
shall only sketch some points of the proof. Let us start with a preliminary
lemma.
Lemma 6.9 Let f ∈ C
B
([0, T]; L
2
(Ω)) and let i, j ∈ ¦1, 2..., m¦. Then we
have
lim
σ→0
n
k=1
f(t
k−1
)(B
i
(t
k
) −B
i
(t
k−1
))(B
j
(t
k
) −B
j
(t
k−1
))
= δ
i,j
_
T
0
f(s)ds, in L
2
(Ω, F, P).
(6.29)
Proof. Let η = ¦0 = t
0
< t
1
< < t
n
= T¦ be a decomposition of [0, T].
If i = j, (6.29) follows from Lemma 6.2. Let i ,= j and set
I
η
i,j
:=
n
k=1
f(t
k−1
)(B
i
(t
k
) −B
i
(t
k−1
))(B
j
(t
k
) −B
j
(t
k−1
)).
Then we have
E[(I
σ
i,j
)
2
] = E
n
h,k=1
f(t
h−1
)f(t
k−1
)(B
i
(t
h
) −B
i
(t
h−1
))(B
j
(t
h
) −B
j
(t
h−1
))
(B
i
(t
k
) −B
i
(t
k−1
))(B
j
(t
k
) −B
j
(t
k−1
))
= E
n
h=1
f
2
(t
h−1
)(B
i
(t
h
) −B
i
(t
h−1
))
2
(B
j
(t
h
) −B
j
(t
h−1
))
2
=
n
h=1
E(f
2
(t
h−1
))(t
h
−t
h−1
)
2
→ 0,
86 Chapter 6
as [σ[ → 0.
Now we prove Itˆo’s formula when b and σ are elementary processes as,
b =
p
i=1
b
i−1
1l
[λ
i−1
,λ
i
)
, σ =
p
i=1
σ
i−1
1l
[λ
i−1
,λ
i
)
, (6.30)
where p ∈ N, 0 = λ
0
< λ
1
< < λ
p
b
i
∈ L
2
(Ω, F
t
i
, P; R
d
) and σ
i
∈
L
2
(Ω, F
t
i
, P; L(R
m
; R
d
)) i = 0, 1, ..., p −1.
Lemma 6.10 Let ϕ ∈ C
2
b
(R
d
), x ∈ R
d
and let b and σ given by (6.30).
Then identity (6.27) holds.
Proof. We proceed as in the proof of Lemma 6.3, taking ϕ ∈ C
3
b
(R
d
) and
proving (6.6) in [0, t] with t ≤ λ
1
. We have
b(t) = b
0
, σ(t) = σ
0
, t ∈ [0, λ
1
]
and
X(t) = b
0
t + σ
0
B(t), t ∈ [0, λ
1
].
Let η = ¦t
0
= 0 < t
1
< < t
N
= t¦. Then we obviously have
ϕ(X(t)) −ϕ(x) =
N
k=1
[ϕ(X(t
k
)) −ϕ(X(t
k−1
))].
On the other hand, by Taylor’s formula we can write
(3)
ϕ(X(t)) −ϕ(x) =
N
k=1
¸Dϕ(X(t
k−1
)), X(t
k
) −X(t
k−1
))
+
1
2
N
k=1
¸D
2
ϕ(X(t
k−1
))(X(t
k
) −X(t
k−1
)), X(t
k
) −X(t
k−1
)) + R
η
=: I
1
+ I
2
+ I
3
. (6.31)
Concerning I
1
we have
I
1
=
N
k=1
¸Dϕ(X(t
k−1
)), b
0
(t
k
−t
k−1
) + σ
0
(B(t
k
) −B(t
k−1
)).
(3)
We use the notations Dϕ(x)h = ¸Dϕ(x), h) and D
2
ϕ(x)(h, k) = ¸D
2
ϕ(x)h, k) for all
x, h, k ∈ R
d
.
The Itˆo formula 87
So,
lim
η→0
I
1
=
_
t
0
¸Dϕ(X(s)), b(s))ds+
_
t
0
¸Dϕ(X(s)), σ(s)dB(s)) in L
2
(Ω, F, P).
(6.32)
Concerning I
2
we write
2I
2
=
N
k=1
¸D
2
ϕ(X(t
k−1
))b
0
, b
0
)(t
k
−t
k−1
)
2
+ 2
N
k=1
¸D
2
ϕ(X(t
k−1
))b
0
, σ
0
(B(t
k
) −B(t
k−1
)))(t
k
−t
k−1
)
+
N
k=1
¸D
2
ϕ(X(t
k−1
))σ
0
(B(t
k
)−B(t
k−1
)), σ
0
(B(t
k
)−B(t
k−1
))) =: I
2,1
+I
2,2
+I
2,3
.
(6.33)
It is easy to check that
lim
η→0
I
2,1
= lim
η→0
I
2,2
= 0 in L
1
(Ω, F, P) (6.34)
Moreover, we have
2I
2,3
=
N
k=1
¸D
2
ϕ(X(t
k−1
))(σ(B(t
k
) −B(t
k−1
))), σ(B(t
k
) −B(t
k−1
)))
=
N
k=1
d
i,j=1
m
α,β=1
D
2
i,j
ϕσ
i,α
(B
α
(t
k
) −B
α
(t
k−1
)) σ
i,β
(B
β
(t
k
) −B
β
(t
k−1
)).
Therefore, taking into account Lemma 6.9 we have
lim
η→0
2I
2,3
=
_
t
0
d
i,j=1
m
α=1
D
2
i,j
ϕ(X(s)) σ
i,α
(s)σ
i,β
(s)ds
=
_
t
0
Tr [D
2
ϕ(X(s))(σσ
∗
(s))]ds.
Now, proceeding as before, we see that
lim
η→0
E[R
η
[ = 0, (6.35)
88 Chapter 6
The proof is complete when t ≤ λ
1
. The general case can be treated in
the same way taking into account that b
k−1
and σ
k−1
are independent of
B(t
k
) −B(t
k−1
).
Finally, proceeding as we did for the proof of Theorem 6.4 we obtain the
result
Theorem 6.11 Let b ∈ L
2
([0, T] Ω, P, dt P : R
d
), σ ∈ L
2
([0, T]
Ω, P, dt P : L(R
m
; R
d
)), x ∈ R
d
and ϕ ∈ C
2
b
(R
d
). Then identity (6.27)
holds for any t ∈ [0, T].
Exercise 6.12 Let d = 1, m ∈ N, b, σ
k
∈ L
2
([0, T] Ω, P, dt P), k =
1, ..., m.
Set
X(t) =
_
t
0
b(s)ds +
m
k=1
_
t
0
σ
k
(s)dB
k
(s).
Let ϕ ∈ C
2
b
(R). Prove that
dϕ(X(t)) = ϕ
(X(t))dX(t) +
1
2
ϕ
(X(t))[σ(t)[
2
dt, (6.36)
where σ(t) = (σ
1
(t), ..., σ
m
(t)).
Exercise 6.13 Let d ∈ N, m = 1 b
i
, σ
i
∈ L
2
([0, T] Ω, P, dt P), i = 1, 2 =
..., d. Set
X(t) = b(t)dt + σdB(t), i = 1, 2,
where σ = (σ
1
, ..., σ
d
). Let moreover ϕ ∈ C
2
b
(R
d
). Prove that
dϕ(X(t)) = ¸Dϕ(X(t)), dX(t)) +
1
2
¸D
2
ϕ(X(t))σ(t), σ(t))dt. (6.37)
Chapter 7
Stochastic evolution equations
We are given two positive integers r, d and an rdimensional standard Brow
nian motion B(t), t ≥ 0, in a probability space (Ω, F, P). We denote by
(F
t
)
t≥0
the natural ﬁltration of B(t) (augmented with all Pnull sets of Ω).
Let us consider the following integral equation
X(t) = η +
_
t
s
b(u, X(u))du +
_
t
s
σ(u, X(u))dB(u), t ∈ [s, T], (7.1)
where s ∈ [0, T), η ∈ L
2
(Ω, F
s
, P; R
d
), b: [0, T] R
d
→ R
d
and σ: [0, T]
R
d
→ L(R
r
, R
d
). b is called the drift and σ the diﬀusion coeﬃcient of the
equation.
We shall write (7.1) in diﬀerential form as
_
_
_
dX(t) = b(t, X(t))dt + σ(t, X(t))dB(t),
X(s) = η.
(7.2)
By a solution of equation (7.1) on the interval [s, T] we mean a function
X ∈ C
B
([s, T]; L
2
(Ω; R
d
)) that fulﬁlls equation (7.1).
In order to solve (7.1) we shall use a ﬁxed point argument, based on the
identity
E
¸
¸
¸
¸
_
b
a
G(t)dB(t)
¸
¸
¸
¸
2
=
_
b
a
E[Tr (G(t)G
∗
(t))] dt.
for all G ∈ C
B
([0, T]; L
2
(Ω, L(R
r
, R
d
))) and 0 ≤ a < b ≤ T. This suggests to
endow L(R
r
, R
d
) with the Hilbert–Schmidt norm, setting
S
HS
: = [Tr(SS
∗
)]
1/2
, S ∈ L(R
r
, R
d
)
and to write
E
¸
¸
¸
¸
_
b
a
G(t)dB(t)
¸
¸
¸
¸
2
=
_
b
a
E
_
G(t)
2
HS
_
dt. (7.3)
89
90 Chapter 7
7.1 Existence and uniqueness
The standard assumptions for the wellposedness of problem (7.1) are the
following.
Hypothesis 7.1
(i) b and σ are continuous on [0, T] R
d
.
(ii) There exists M > 0 such that for all t ∈ [0, T], x, y ∈ R
d
, we have
[b(t, x) −b(t, y)[
2
+σ(t, x) −σ(t, y)
2
HS
≤ M
2
[x −y[
2
(7.4)
and
[b(t, x)[
2
+σ(t, x)
2
HS
≤ M
2
(1 +[x[
2
). (7.5)
Notice that, after possibly changing the constant M, (7.5) is a consequence
of (7.4).
Theorem 7.1 Assume that Hypothesis 7.1 holds and let s ∈ [0, T), η ∈
L
2
(Ω, F
s
, P; R
d
). Then problem (7.1) has a unique solution
X ∈ C
B
([s, T]; L
2
(Ω; R
d
)).
Proof. We are going to solve (7.1) by a ﬁxed point argument in the space
C
B
:= C
B
([s, T]; L
2
(Ω; R
d
)).
Deﬁne
γ
1
(X)(t) :=
_
t
s
b(u, X(u))du, X ∈ C
B
, t ∈ [s, T],
γ
2
(X)(t) :=
_
t
s
σ(u, X(u))dB(u), X ∈ C
B
, t ∈ [s, T]
and set
γ(X) := η + γ
1
(X) + γ
2
(X), X ∈ C
B
.
Then equation (7.1) is equivalent to the following,
X = η + γ
1
(X) + γ
2
(X) = γ(X). (7.6)
Step 1. γ
1
and γ
2
map C
B
into itself.
Stochastic evolution equations 91
Concerning γ
1
we have, using the H¨ older inequality and taking into ac
count (7.5),
[γ
1
(X)(t)[
2
≤ (t −s)
_
t
s
[b(u, X(u))[
2
du ≤ M
2
(t −s)
_
t
s
(1 +[X(u)[
2
)du
≤ M
2
(t −s)
2
(1 +X
2
C
B
).
Since γ
1
(X)(t) is F
t
–measurable for all t ∈ [s, T], γ
1
maps C
B
into itself and
γ
1
(X)
C
B
≤ M(t −s)(1 +X
C
B
).
Concerning γ
2
we have taking into account (7.3) and (7.5),
E[γ
2
(X)(t)[
2
=
_
t
s
E(σ(u, X(u))
2
HS
)du
≤ M
2
_
t
s
(1 +[X(u)[
2
)du ≤ M
2
(t −s)(1 +X
2
C
B
)
So, we see that γ
2
maps C
B
into itself.
Step 2. γ is Lipschitz continuous.
Let X, Y ∈ C
B
. We have, using again the H¨older inequality and taking
into account (7.4),
[γ
1
(X)(t) −γ
1
(Y )(t)[
2
≤ (t −s)
_
t
s
[b(u, X(u)) −b(u, Y (u))[
2
du
≤ (t −s)M
2
_
t
s
[X(u) −Y (u)[
2
du ≤ (t −s)
2
M
2
X −Y 
2
C
B
du.
Consequently
γ
1
(X) −γ
1
(Y )
C
B
≤ M (T −s) X −Y 
C
B
, X, Y ∈ C
B
(7.7)
Furthermore
E[γ
2
(X)(t) −γ
2
(Y )(t)[
2
=
_
t
s
E(σ(u, X(u)) −σ(u, Y (u))
2
HS
)du
≤ M
2
(t −s)X −Y 
2
C
B
,
92 Chapter 7
and so,
γ
2
(X) −γ
2
(Y )
C
B
≤ M
√
T −s X −Y 
C
B
, X, Y ∈ C
B
. (7.8)
By (7.7) and (7.8) it follows that γ maps C
B
into itself and
γ(X) −γ(Y )
C
B
≤ M(T −s +
√
T −s )X −Y [
C
B
,
for all X, Y ∈ C
B
. Now if T −s is such that
M
_
T −s +
√
T −s
_
≤ 1/2, (7.9)
γ is a 1/2–contraction on C
B
, and so, it possesses a unique ﬁxed point. If
(7.9) does not hold we choose T
1
∈ (s, T] such that
M
_
T
1
−s +
_
T
1
−s
_
≤ 1/2.
Then by the previous argument there is a unique solution to (7.1) on [s, T
1
].
Now we repeat the proof with T
1
replacing s and in a ﬁnite number of steps
we arrive to the conclusion.
Remark 7.2 By Theorem 5.13 it follows that there exists a version of the
solution X(, s, η) which belongs to L
2
(Ω, C([s, T])) and so it is a continuous
process.
In the following we shall denote by X(, s, η) the solution of problem (7.1).
Whe shall use greek letters for stochastic initial data and latin letters for
deterministic ones.
Let us prove the cocycle law.
Proposition 7.3 Assume that Hypothesis 7.1 holds and let η ∈ L
2
(Ω, F
s
, P; R
d
).
Then
X(t, s, η) = X(t, r, X(r, s, η)), 0 ≤ s ≤ r ≤ t ≤ T. (7.10)
Proof. Deﬁne Z(t) = X(t, s, η), t ∈ [s, T]. Then Z solves the problem
_
_
_
dZ(t) = b(t, Z(t))dt + σ(t, Z(t))dB(t),
Z(r) = X(r, s, η).
By the uniqueness part of Theorem 7.1 it follows that
Z(t) = X(t, s, η) = X(t, r, X(r, s, η)),
as required.
Stochastic evolution equations 93
Remark 7.4 By the contraction principle it follows that the solution X(t, s, η)
of problem (7.1) can be obtained as a limit of successive approximations.
More precisely, deﬁne X
0
(t, s, η) = η and for any N ∈ N,
X
N+1
(t, s, η) = η +
_
t
s
b(u, X
N
(u, s, η))du +
_
t
s
σ(u, X
N
(u, s, η))dB(u).
(7.11)
Then we have
lim
N→∞
X
N
(, s, η) = X(, s, η) in C
B
([s, T]; L
2
(Ω; R
d
)). (7.12)
Next result, which as we shall see plays an important rˆole in proving that
X(, s, x) is a Markov process, gives some information about the relationship
between X(t, s, η), η ∈ L
2
(Ω, F
s
, P; R
d
) and X(t, s, x), x ∈ R
d
.
Proposition 7.5 Assume that Hypothesis 7.1 holds and that
η =
n
k=1
x
k
1l
A
k
, (7.13)
where x
1
, ..., x
n
∈ R
d
, and A
1
, ..., A
n
are mutually disjoints sets in F
s
such
that
Ω =
n
_
k=1
A
k
.
Then we have
X(t, s, η) =
n
k=1
X(t, s, x
k
)1l
A
k
. (7.14)
Proof. Let X
N
be deﬁned by (7.11). We claim that
X
N
(t, s, η) =
n
k=1
X
N
(t, s, x
k
)1l
A
k
, ∀ N ∈ N. (7.15)
Once (7.15) is proved, the conclusion follows letting N tend to inﬁnity. Let
us proceed by recurrence. Equality (7.15) is clear for N = 0. Assume that it
holds for a given N ∈ N, so that
X
N
(t, s, η) = X
N
(t, s, x
k
) in A
k
, k = 1, ..., n.
Then we have
b(u, X
N
(u, s, η)) = b(u, X
N
(u, s, x
k
)) in A
k
, k = 1, ..., n,
σ(u, X
N
(u, s, η)) = σ(u, X
N
(u, s, x
k
)) in A
k
, k = 1, ..., n,
94 Chapter 7
so that
b(u, X
N
(u, s, η)) =
n
k=1
1l
A
k
b(u, X
N
(u, s, x
k
)),
σ(u, X
N
(u, s, η)) =
n
k=1
1l
A
k
σ(u, X
N
(u, s, x
k
)).
Consequently
X
N+1
(t, s, η) =
n
k=1
1l
A
k
_
X
0
(t, s, x
k
) +
_
t
s
b(u, X
N
(u, s, x
k
)du
+
_
t
s
σ(u, X
N
(u, s, x
k
))dB(u)
_
=
n
k=1
1l
A
k
X
N+1
(t, s, x
k
)
and (7.15) holds for N + 1. So, the conclusion follows.
7.1.1 Solution of the stochastic diﬀerential equation in
the space C
B
([s, T]; L
2m
(Ω; R
d
)).
Theorem 7.6 Assume that Hypothesis 7.1 holds and let m ∈ N, s ∈ [0, T),
η ∈ L
2m
(Ω, F
s
, P; R
d
). Then problem (7.1) has a unique solution
X(, s, η) ∈ C
B
([s, T]; L
2m
(Ω; R
d
)).
In particular
X(, s, x) ∈ C
B
([s, T]; L
2m
(Ω; R
d
)), ∀ x ∈ R
d
.
Proof. We proceed as in the proof of Theorem 7.1 by a ﬁxed point argument
in the space
C
m
B
:= C
B
([s, T]; L
2m
(Ω; R
d
)),
using inequality (6.24) proved in Proposition 6.8.
7.1.2 Examples
Example 7.7 Consider the stochastic diﬀerential equation
dX = AXdt + CdB(t), X(0) = x, (7.16)
where A ∈ L(R
d
), C ∈ L(R
r
; R
d
) and x ∈ R
d
.
Stochastic evolution equations 95
Clearly Theorem 7.1 applies so that (7.16) has a unique solution X(t)
which fulﬁlls the integral equation
X(t) = x + A
_
t
0
X(s)ds + CB(t). (7.17)
Setting
Y (t) =
_
t
0
X(s)ds, t ∈ [0, T],
Y fulﬁlls the equation
Y
(t) = AY (t) + x + CB(t), Y (0) = 0, t ∈ [0, T],
which can be easily solved by the method of variation of constants. We
obtain
Y (t) =
_
t
0
e
(t−s)A
(x + CB(s))ds, t ∈ [0, T].
By substituting Y (t) in (7.17) yields
X(t) = A
_
t
0
e
(t−s)A
(x + CB(s))ds + x + CB(t).
Taking into account that, thanks to Proposition 3.12,
_
t
0
e
(t−s)A
CdB(s) = CB(t) + A
_
t
0
e
(t−s)A
CB(s)ds,
we ﬁnd
X(t) = e
tA
x +
_
t
0
e
(t−s)A
CdB(s). (7.18)
Example 7.8 Let r = d = 1 and consider the stochastic diﬀerential equation
dX = aXdt + cXdB(t), X(0) = x, (7.19)
where a, c, x ∈ R. Again Theorem 7.1 applies. We want to show that the
solution of (7.19) is given by
X(t) = e
t
(
a−
1
2
c
2
)
e
cB(t)
x, t ≥ 0. (7.20)
For this we check that X(t) given by (7.20) solves (7.19).
Write X(t) = e
F(t)
where F(t) = t
_
a −
1
2
c
2
_
+ cB(t). Then we have
dF(t) =
_
a −
1
2
c
2
_
dt + cdB(t)
96 Chapter 7
and, by Itˆo’s formula,
dX(t) = e
F(t)
dF(t) +
1
2
c
2
e
F(t)
dt
= e
F(t)
_
a −
1
2
c
2
_
dt + cdB(t) +
1
2
c
2
e
F(t)
dt
= aX(t)dt + cX(t)dB(t).
Exercise 7.9 Let r = 1 and consider the diﬀerential stochastic equation
dX = AXdt + CXdB(t), X(0) = x, (7.21)
where A, C ∈ L(R
d
), x ∈ R
d
and AC = CA. Show that the solution of (7.21)
is given by
X(t) = e
t(A−C
2
/2)
e
CB(t)
x. (7.22)
7.1.3 Diﬀerential stochastic equations with random co
eﬃcients
In some situations (see Subsections 7.3 and 7.4) one deals with stochastic
diﬀerential equations having random coeﬃcients,
X(t, ω) = η(ω) +
_
t
s
b(u, X(u, ω), ω)du +
_
t
s
σ(u, X(u, ω), ω)dB(u). (7.23)
Here η ∈ L
2
(Ω, F
s
, R
d
), b: [0, T] R
d
Ω →R
d
and σ: [0, T] L(R
r
, R
d
)
Ω →R
d
are such that:
Hypothesis 7.2
(i) There exists M > 0 such that for all t ∈ [0, T], x, y ∈ R
d
, ω ∈ Ω
[b(t, x, ω)−b(t, y, ω)[
2
+σ(t, x, ω)−σ(t, y, ω)
2
HS
≤ M
2
[x−y[
2
(7.24)
and
[b(t, x, ω)[
2
+σ(t, x, ω)
2
HS
≤ M
2
(1 +[x[
2
). (7.25)
(ii) For any Y ∈ C
B
([0, T]; L
2
(Ω, R
d
)) we have U ∈ C
B
([0, T]; L
2
(Ω, R
d
))
and V ∈ C
B
([0, T]; L
2
(Ω, L(R
r
, R
d
))) where, for all t ∈ [0, T], ω ∈ Ω,
U(t, ω) = b(t, Y (t, ω), ω)), V (t, ω) = σ(t, Y (t, ω), ω)).
The following result can be proved as Theorem 7.1.
Stochastic evolution equations 97
Theorem 7.10 Assume that Hypothesis 7.2 holds. Let s ∈ [0, T) and η ∈
L
2
(Ω, F
s
, R
d
). Then problem (7.23) has a unique solution
X ∈ C
B
([s, T]; L
2
(Ω; R
d
)).
Example 7.11 Let d = 1 and consider the stochastic diﬀerential equation
_
_
_
dX(t) = X(t)¸F(t), dB(t)), t ∈ [0, T],
X(0) = x,
(7.26)
where F ∈ C
B
(0, T; L
∞
(Ω; R
d
)). Now it is easy to check that Theorem 7.10
applies and so there exists a solution X of (7.26). Let us show that
X(t) = e
−
1
2
R
t
0
F(s)
2
ds+
R
t
0
F(s),dB(s)
x, t ≥ 0. (7.27)
For this we check that X(t) given by (7.27) solves (7.26).
Write X(t) = e
H(t)
where
H(t) = −
1
2
_
t
0
[F(s)[
2
ds +
_
t
0
¸F(s), dB(s)).
Then we have
dH(t) = −
1
2
[F(t)[
2
dt +¸F(t), dB(t)), t ≥ 0.
Now by Itˆo’s formula we ﬁnd
dX(t) = e
H(t)
dH(t) +
1
2
e
H(t)
[F(t)[
2
dt
= e
H(t)
¸F(t), dB(t)) = X(t)¸F(t), dB(t)), t ≥ 0.
So, (7.27) is proved.
7.2 Continuous dependence on data
7.2.1 Continuous dependence on mean square
We assume here that Hypothesis 7.1 holds. We are going to prove that
the solution X(t, s, η) to (7.1) is H¨ older continuous on t, s and Lipschitz
continuous on η in mean square. First we show that E[X(t, s, η)[
2
is bounded.
98 Chapter 7
Lemma 7.12 Assume that Hypothesis 7.1 holds. Then for all s ∈ [0, T] and
η ∈ L
2
(Ω, F
s
, P; R
d
) we have
E
_
[X(t, s, η)[
2
_
≤ 3[E([η[
2
) + M
2
((T −s)
2
+ (T −s)]e
3M
2
(T−s+1)
. (7.28)
Proof. Writing for short X(t, s, η) = X(t), we have
E([X(t)[
2
) ≤ 3E([η[
2
) + 3E
_
¸
¸
¸
¸
_
t
s
b(u, X(u))du
¸
¸
¸
¸
2
_
+3
_
t
s
E(σ(u, X(u))
2
HS
)du.
By Hypothesis 7.1(ii) and the H¨ older inequality we deduce that
E([X(t)[
2
) ≤ 3E([η[
2
) + 3M
2
(t −s)
_
t
s
(1 +E
_
[X(u)[
2
_
)du
+3M
2
_
t
s
(1 +E
_
[X(u)[
2
_
)du.
Consequently
E([X(t)[
2
) ≤ 3E([η[
2
) + 3M
2
((T −s)
2
+ (T −s))
+3M
2
((T −s) + 1)
_
t
s
E
_
[X(u)[
2
_
du.
The conclusion follows from the Gronwall lemma.
We now study the regularity of X(t, s, η) with respect to t, s, η. We note
that, by Lemma 7.12, there exists a constant C(T, E([η[
2
)) such that
E
_
[X(t, s, η)[
2
_
≤ C(T, E([η[
2
)), 0 ≤ s < t ≤ T. (7.29)
We start with the regularity of X(t, s, η) with respect to t.
Proposition 7.13 Assume that Hypothesis 7.1 holds. Let 0 ≤ s ≤ t
1
< t ≤
T and η ∈ L
2
(Ω, F
s
, R
d
). Then there exists a constant C
1
(T, E([η[
2
)) such
that we have
E
_
[X(t, s, η) −X(t
1
, s, η)[
2
_
≤ C
1
(T, E([η[
2
))(t −t
1
). (7.30)
Stochastic evolution equations 99
Proof. We have
E
_
[X(t, s, η) −X(t
1
, s, η)[
2
_
≤ 2M
2
(t −t
1
)
_
t
t
1
(1 +E
_
[X(u, s, η)[
2
_
du
+ 2M
2
_
t
t
1
(1 +E
_
[X(u, s, η)[
2
_
)du.
Consequently,
E
_
[X(t, s, η) −X(t
1
, s, η)[
2
_
≤ 2M
2
((t −t
1
)
2
+ t −t
1
)(1 + C
2
(T, E([η[
2
)))
and the conclusion follows.
Let us study the regularity of X(t, s, η) with respect to η.
Proposition 7.14 Assume that Hypothesis 7.1 holds, let 0 ≤ s < t ≤ T and
η, ζ ∈ L
2
(Ω, F
s
, R
d
). Then
E
_
[X(t, s, η) −X(t, s, ζ)[
2
_
≤ 3e
3M
2
(T−s+1)(t−s)
E([η −ζ[
2
). (7.31)
Proof. We have
[X(t, s, η) −X(t, s, ζ)[
2
≤ 3[η −ζ[
2
+ 3
¸
¸
¸
¸
_
t
s
(b(u, X(u, s, η) −b(u, X(u, s, ζ))du
¸
¸
¸
¸
2
+ 3
¸
¸
¸
¸
_
t
s
(σ(u, X(u, s, η) −σ(u, X(u, s, ζ))dB(u)
¸
¸
¸
¸
2
.
Taking expectation and using (7.4) we obtain
E([X(t, s, η) −X(t, s, ζ)[
2
) ≤ 3E([η −ζ[
2
) + 3M
2
(T −s + 1)
_
t
s
E
_
[X(u, s, η) −X(u, s, ζ)[
2
_
du
and the conclusion follows from the Gronwall lemma.
We ﬁnally study the regularity of X(t, s, η) with respect to s.
Proposition 7.15 Assume that Hypothesis 7.1 holds, let 0 < s < s
1
< t ≤
T, and η ∈ L
2
(Ω, F
s
, P; R
d
). Then there exists a constant C
T,η
> 0 such that
E
_
[X(t, s, η) −X(t, s
1
, η)[
2
_
≤ C
T,η
[s −s
1
[. (7.32)
100 Chapter 7
Proof. Taking into account the cocycle law (7.10), we can write
X(t, s, η) −X(t, s
1
, η) = X(t, s
1
, X(s
1
, s, η)) −X(t, s
1
, η).
By (7.31) there exists C
T
> 0 such that
E([X(t, s, η) −X(t, s
1
, η)[
2
) ≤ C
2
T
E([X(s
1
, s, η) −η[
2
)
= C
2
T
E([X(s
1
, s, η) −X(s, s, η)[
2
) .
The conclusion follows now from (7.30).
7.3 Almost sure continuity and h¨olderianity
of trajectories
In this section we show that X(, s, x) belongs to a suitable Sobolev space,
whose deﬁnition is recalled in Appendix E below. Then the Sobolev embed
ding theorem (also stated in Appendix E) will imply that X(, s, x) is H¨ older
continuous almost surely.
First we need a lemma, which can be proved as Proposition 7.13 using
(6.24).
Lemma 7.16 Assume that Hypothesis 7.1 holds. Let 0 ≤ s ≤ t
1
< t ≤
T, x ∈ R
d
and m ∈ N. Then there exists a constant C
1
(T, [x[) such that
E
_
[X(t, s, x) −X(t
1
, s, x)[
2m
_
≤ C
1
(T, [x[
2
))(t −t
1
)
m
. (7.33)
Now from Proposition E.3 and the Sobolev embedding theorem E.1 it
follows that
Proposition 7.17 Assume that Hypothesis 7.1 holds. Let x ∈ R
d
, 0 ≤ s ≤
t ≤ T, m ∈ N and ∈ (0, 1/2). Then we have
E
_
[X(, s, x)[
2m
,2m
¸
< +∞. (7.34)
Moreover, X(, s, x) belongs to C
−1/(2m)
([s, T]) almost surely.
Finally, we consider almost sure regularity of X(t, s, ). First, arguing as
in the proof of Proposition 7.14 we have
Lemma 7.18 Assume that Hypothesis 7.1 holds, let 0 ≤ s < t ≤ T and
x, y ∈ R
d
. Then there is a constant C(T) > 0 such that
E
_
[X(t, s, x) −X(t, s, y)[
2m
_
≤ C(T)[x −y[
2m
. (7.35)
Stochastic evolution equations 101
Now from Proposition E.3 it follows that
Proposition 7.19 Assume that Hypothesis 7.1 holds, let 0 ≤ s < t ≤ T and
x, y ∈ [0, 1]
d
. Then for any m > 1 and ∈ (0, 1) we have
E
_
[X(t, s, )[
2m
,2m
¸
< +∞. (7.36)
Moreover, X(t, s, ) belongs to C
−d/(2m)
([0, 1]
d
) almost surely.
7.4 Diﬀerentiability of X(t, s, x) with respect
to x
In this section we assume, besides Hypothesis 7.1, that
Hypothesis 7.3
(i) D
x
b, D
2
x
b, D
x
σ and D
2
x
σ are continuous on [0, T] R
d
.
(ii) We have
(1)
sup
t∈[0,T]
([b(t, )]
2
+ [σ(t, )]
2
) < ∞. (7.37)
We set
C
B
= C
B
([s, T]) =: C
B
([s, T]; L
2
(Ω; R
d
)).
7.4.1 Existence of X
x
(t, s, x)
Theorem 7.20 Assume that Hypotheses 7.1 and 7.3 hold. Then for any
s ∈ [0, T] the mapping
R
d
→ C
B
, x → X(, s, x),
is continuously Gateaux diﬀerentiable and its Gateaux derivative is given by
X
x
(t, s, x) h = η
h
(t, s, x), x, h ∈ R
d
, (7.38)
where η
h
(t, s, x) is the solution to the stochastic diﬀerential equation with
random coeﬃcients,
_
¸
¸
¸
¸
_
¸
¸
¸
¸
_
dη
h
(t, s, x) = b
x
(t, X(t, s, x)) η
h
(t, s, x)dt
+σ
x
(t, X(t, s, x))(η
h
(t, s, x), dB(t))
η
h
(s, s, x) = h.
(7.39)
(1)
Recall the notations given at the beginning of Chapter 6.
102 Chapter 7
Proof. Note that the coeﬃcients of equation (7.39) fulﬁll Hypothesis 7.3, so
it possesses a unique solution by Theorem 7.10.
To prove the theorem we use Theorem D.6 from Appendix D (with Λ = R
d
and E = C
B
). We set C
B
= C
B
([s, T
1
]) and deﬁne a mapping
F : R
d
C
B
→ C
B
,
setting
[F(x, X)](t): = x +
_
t
s
b(r, X(r))dr +
_
t
s
σ(r, X(r))dB(r), t ∈ [s, T
1
],
(7.40)
where T
1
> s is chosen such that
F(x, X
1
) −F(x, X
2
)
C
B
≤
1
2
X
1
−X
2

C
B
for all X
1
, X
2
∈ C
B
, x ∈ R
d
.
(7.41)
Then F fulﬁlls Hypothesis D.1 so that it possesses a unique ﬁxed point
X(x) ∈ C
B
, that is
F(x, X(x)) = X(x), x ∈ R
d
,
which depends continuously on x. X(x) coincides with the solution X(, s, x)
of (7.2).
It is not diﬃcult to check that F is Gateaux continuously diﬀerentiable,
(the straightforward proof is left to the reader) and that for each x ∈ R
d
,
X, Y ∈ C
B
we have
F
x
(x, X) = I,
[F
X
(x, X)Y ](t) =
_
t
s
b
x
(r, X(r))Y (r)dr+
_
t
s
σ
x
(r, X(r))Y (r)dB(r), t ∈ [s, T
1
],
So, the conclusion follows from Theorem D.6.
7.4.2 Existence of X
xx
(t, s, x)
We now prove the existence of the second derivative of X(t, s, x) with respect
to x.
Theorem 7.21 Assume that Hypotheses 7.1 and 7.3 hold. Then the mapping
R
d
→ C
B
, x → X(, s, x),
is twice diﬀerentiable with respect to x in any couple of directions (h, k) in
R
d
. Moreover, setting
X
xx
(t, s, x)(h, k) = ζ
h,k
(t, s, x), x, h ∈ R
d
, (7.42)
Stochastic evolution equations 103
ζ
h,k
(t, s, x) is the solution to the stochastic diﬀerential equation (with random
coeﬃcients)
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
d ζ
h,k
(t, s, x) = b
x
(t, X(t, s, x)) ζ
h,k
(t, s, x)dt
+b
xx
(t, X(t, s, x))(η
h
(t, s, x), η
k
(t, s, x))dt
+σ
x
(t, X(t, s, x))(ζ
h,k
(t, s, x), dB(s))
+σ
xx
(t, X(t, s, x)) (η
h
(t, s, x), η
k
(t, s, x), dB(t))
ζ
h,k
(s, s, x) = 0.
(7.43)
We shall prove the theorem when n = r = 1 for simplicity. We ﬁrst prove a
lemma.
Lemma 7.22 Let η(, s, x) ∈ C
B
([s, T]; L
2
(Ω)) be the solution of the equa
tion
η(t, s, x) = 1 +
_
t
s
b
x
(r, X(r, s, x))η(r, s, x)dr
+
_
t
s
σ
x
(r, X(r, s, x))η(r, s, x)dB(r).
(7.44)
Then η(, s, x) ∈ C
B
([s, T]; L
4
(Ω)) and there exists C > 0 such that
E[η(, s, x)[
4
≤ C, ∀ s ∈ [0, T), x ∈ R
d
. (7.45)
Proof. We have,
[η(t, s, x)[
4
≤ 27 + 27
¸
¸
¸
¸
_
t
s
b
x
(r, X(r, s, x))η(r, s, x)dr
¸
¸
¸
¸
4
+27
¸
¸
¸
¸
_
t
s
σ
x
(r, X(r, s, x))η(r, s, x)dB(r)
¸
¸
¸
¸
4
.
By using (7.37) and the H¨older inequality we see that there exists a constant
C
1
such that
[η(t, s, x)[
4
≤ 27 + C
1
_
t
s
[η(r, s, x)[
4
dr
+C
1
¸
¸
¸
¸
_
t
s
σ
x
(r, X(r, s, x))η(r, s, x)dB(r)
¸
¸
¸
¸
4
.
104 Chapter 7
Now, taking expectation on both sides of this inequality and using Corollary
6.8, we ﬁnd that
E[η(t, s, x)[
4
≤ C
2
(1 +
_
t
s
E[η(r, s, x)[
4
dr), 0 ≤ s ≤ t ≤ T, x ∈ R,
where C
2
is another constant. The conclusion follows from the Gronwall
lemma.
Proof of Theorem 7.21. We choose T
1
as in (7.41) and C
B
= C
B
([s, T
1
])
as before. By Theorem 7.20 we know that X(t, s, x) is diﬀerentiable with
respect to x and that its derivative η(, s, x) = X
x
(, s, x) belongs to C
B
and
fulﬁlls equation (7.44). For any x ∈ R we deﬁne a linear bounded operator
T(x) from C
B
into C
B
setting for all t ∈ [s, T
1
],
(T(x)Z)(t) = −
_
t
s
b
x
(r, X(r, s, x))Z(r)dr −
_
t
s
σ
x
(r, X(r, s, x))Z(r)dB(r).
(7.46)
Notice that, since η(, s, x) ∈ C
B
([s, T]; L
4
(Ω)), T(x)Z is diﬀerentiable with
respect to x for any Z ∈ C
B
([s, T]; L
4
(Ω)) and it results
(T
(x)Z)(t) = −
_
t
s
b
xx
(r, X(r, s, x))Z(r)η(, r, x)dr
−
_
t
s
σ
xx
(r, X(r, s, x))Z(r)η(, r, x)dB(r).
(7.47)
Now we write equation (7.44) as
η(, s, x) = 1 + T(x)η(, s, x) (7.48)
By (7.41) it follows that
T(x)
L(C
B
)
≤ 1/2, ∀ x ∈ R.
Thus the solution of (7.48) is given by
η(, s, x) = (1 −T(x))
−1
(1). (7.49)
From this identity it is easy to show the existence of η
x
(, s, x) := ζ(, s, x).
We have in fact, by a straightforward computation
η
x
(, s, x) = (1 −T(x))
−1
(T
(x)η(, s, x)), (7.50)
Stochastic evolution equations 105
where
T
(x)η(, s, x)(t) =
_
t
s
b
xx
(r, X(r, s, x))η
2
(, s, x)dr
+
_
t
s
σ
xx
(r, X(r, s, x))η
2
(, s, x)dB(r).
(7.51)
Now by (7.50) it follows that
η
x
(t, s, x) −T(x)η
x
(, s, x)(t) =
_
t
s
b
xx
(r, X(r, s, x))η
2
(, s, x)dr
+
_
t
s
σ
xx
(r, X(r, s, x))η
2
(, s, x)dB(r),
and the conclusion follows.
7.5 Itˆ o Diﬀerentiability of X(t, s, x) with re
spect to s.
It is useful to recall ﬁrst some results in the deterministic case.
7.5.1 The deterministic case
Let us consider the problem
_
_
_
X
(t) = b(t, X(t)), t ∈ [s, T],
X(s) = x,
(7.52)
under Hypotheses 7.1 and 7.3 with σ = 0. Denote by X(t, s, x) the solution
of (7.52). Let us compute X
s
(t, s, x) (it is well known that X(t, s, x) is C
1
in all variables).
Write
X(t, s, x) = X(t, r, X(r, s, x)), t ≥ r ≥ s. (7.53)
Diﬀerentiating (7.53) with respect to r yields
0 = X
s
(t, r, X(r, s, x)) + X
x
(t, r, X(r, s, x)) X
t
(r, s, x).
Setting r = s we ﬁnd
X
s
(t, s, x) = −X
x
(t, s, x)b(s, x),
106 Chapter 7
which is equivalent to
X(t, s, x) = x +
_
t
s
X
x
(t, r, x)b(r, x)dr, 0 ≤ s ≤ t ≤ T. (7.54)
In the next subsection we are going to generalize this formula for the solution
X(t, s, x) of (7.2).
7.5.2 The stochastic case
Here we want to study the diﬀerentiability of X(t, s, x) with respect to s in
a sense to be precised. A diﬃculty arises since the process s → X(t, s, x) is
not adapted, because X(t, s, x) is not F
s
measurable. It happens, however,
that for any s ∈ [0, T], X(t, s, x) is measurable with respect to the σ–algebra
F
+
s
generated by all sets of the form
¦ω ∈ Ω : (B(s
1
(ω)) −B(s(ω)), ..., B(s
n
(ω)) −B(s(ω))) ∈ A¦ ,
where n ∈ N, 0 ≤ s ≤ s
1
< ... < s
n
≤ T and A ∈ B(R
n
). The family
(F
+
s
)
s∈[0,T]
is called the future ﬁltration of B.
Proposition 7.23 Assume that Hypotheses 7.1 holds. Let x ∈ R
d
, s ∈ [0, T].
Then X(t, s, x) is F
+
s
measurable.
Proof. Let X
N
(t, s, x) be deﬁned by (7.11), N ∈ N. Then X
1
(t, s, x) is
F
+
s
–measurable. We have in fact
X
1
(t, s, x) = x +
_
t
s
b(u, x)du +
_
t
s
σ(u, x)dB(u).
Since
_
t
s
σ(u, x)dB(u) = lim
η→0
n
k=1
σ(t
k−1
, x)(B(t
k
) −B(t
k−1
)),
where η = ¦s = t
0
< t
1
< < t
n
= t¦, then X
1
(t, s, x) is F
+
s
measurable.
We end the proof by recurrence.
Now we introduce the backward Itˆo integral for a process wich is adapted
to the future ﬁltration. For this we need the following result which can be
proved as Lemma 4.3.
Lemma 7.24 Let t
1
< t
2
≤ s, and let ϕ ∈ L
2
(Ω, F
+
s
, P). Then B(t
2
)−B(t
1
)
and ϕ are independent.
Stochastic evolution equations 107
We deﬁne C
B
+([0, T]; L
2
(Ω; L(R
r
; R
d
))) by a straightforward generaliza
tion of the space C
B
([0, T]; L
2
(Ω; L(R
r
; R
d
))) deﬁned in Chapter 5.
The elements of C
B
+([0, T]; L
2
(Ω; L(R
r
; R
d
)) are called stochastic pro
cesses adapted to the future ﬁltration (F
+
t
) and continuous in quadratic
mean.
Let F ∈ C
B
+([0, T]; L
2
(Ω; L(R
r
; R
d
))). For any η ∈ Σ with η = ¦0 =
s
0
< s
1
< < s
n
= T¦ we set
I
σ
(F) =
n
k=1
F(t
k
)(B(t
k
) −B(t
k−1
))
The proof of next theorem is completely similar to that of equation (5.10).
Theorem 7.25 For any F ∈ C
B
+([0, T]; L
2
(Ω; L(R
r
; R
d
))) there exists the
limit
lim
σ→0
I
σ
(F) =:
_
T
0
F(s)dB(s), (7.55)
in L
2
(Ω). Moreover we have
E
_
T
0
F(s)dB(s) = 0, (7.56)
and
E
¸
¸
¸
¸
_
T
0
F(s)dB(s)
¸
¸
¸
¸
2
=
_
T
0
E
_
F(s)
2
HS
¸
ds. (7.57)
_
T
0
F(s)dB(s) is called the backward Itˆo integral of the function F in [0, T].
Exercise 7.26 Let t > s. Prove that
_
t
s
B(r)dB(r) =
1
2
(B(t)
2
−B(s)
2
+ (t −s)).
7.5.3 Backward Itˆo’s formula
Theorem 7.27 Assume that Hypotheses 7.1 and 7.3 hold. Then we have
X(t, s, x) −x =
_
t
s
X
x
(t, r, x) b(r, x)dr
+
1
2
_
t
s
TR [X
xx
(t, r, x)(σ(r, x), σ(r, x))]dr
+
_
t
s
X
x
(t, r, x)(σ(r, x), dB(r))),
(7.58)
108 Chapter 7
where
TR [X
xx
(t, r, x)(σ(r, x), σ(r, x))] =
d
k=1
X
xx
(t, r, x)(σ(r, x)e
k
, σ(r, x)e
k
)
and (e
k
) is any orthonormal basis in R
d
.
Proof. We take d = r = 1 for simplicity. For any η ∈ Σ(s, t) we set
[η[ = max
k=1,...,n
(t
k
−t
k−1
).
If η ∈ Σ(s, t) we have
X(t, s, x) −x = −
n
k=1
[X(t, s
k
, x) −X(t, s
k−1
, x)]
= −
n
k=1
[X(t, s
k
, x) −X(t, s
k
, X(s
k
, s
k−1
, x))]
= −
n
k=1
X
x
(t, s
k
, x)(x −X(s
k
, s
k−1
, x))
−
1
2
n
k=1
X
xx
(t, s
k
, x)(x −X(s
k
, s
k−1
, x))
2
+ o([η[).
(7.59)
Arguing as in the proof of Itˆ o’s formula one can show, after some tedious but
straighforward computations, that
lim
η→0
o([η[) = 0, Pa.s..
On the other hand we have
X(s
k
, s
k−1
, x) −x =
_
s
k
s
k−1
b(r, X(r, s
k−1
, x))dr
+
_
s
k
s
k−1
σ(r, X(r, s
k−1
, x))dB(r)
= b(s
k
, x)(s
k
−s
k−1
) + σ(s
k
, x)(B(s
k
) −B(s
k−1
)) + o(s
k
−s
k−1
).
(7.60)
Stochastic evolution equations 109
(Notice that, since b is deterministic, one can replace in (7.60) b(s
k
, x) with
b(ξ
k
, x) where ξ
k
is any point in [s
k−1
, s
k
].) Substituting (7.60) in (7.59) we
ﬁnd that
X(t, s, x) −x =
n
k=1
X
x
(t, s
k
, x)b(s
k
, x)(s
k
−s
k−1
)
+
n
k=1
X
x
(t, s
k
, x)σ(s
k
, x)(B(s
k
) −B(s
k−1
))
+
1
2
n
k=1
X
xx
(t, s
k
, x)σ
2
(s
k
, x)(B(s
k
) −B(s
k−1
))
2
+I
1
(η) + I
2
(η) + I
3
(η) + o
1
([η[).
(7.61)
Obviously
lim
η→0
I
1
(η) =
_
t
s
X
x
(r, x)b(r, x)dr.
Concerning I
2
(η), we note that it is an integral sum corresponding to the
backward Itˆ o integral since X
x
(t, s
k
, x) is F
+
s
k
measurable by Proposition
7.23. Therefore we have
lim
η→0
I
2
(η) =
_
t
s
X
x
(r, x)σ(r, x)dB(r).
The other terms I
3
(η) and o
1
([η[) can be handled as in the proof of Itˆo’s
formula.
In a similar way one can prove the following backward Itˆo formula.
Theorem 7.28 Let ϕ ∈ C
2
b
(R
d
). Then for any 0 ≤ s < t ≤ T, we have
ϕ(X(t, s, x)) −ϕ(x) =
_
t
s
¸D
x
[ϕ(X(t, r, x))], b(r, x))dr
+
1
2
_
t
s
Tr [D
2
x
[ϕ(X(t, r, x))]σ(r, x)σ
∗
(r, x)]dr
+
_
t
s
¸D
x
[ϕ(X(t, r, x))], σ(r, x)dB(r).
(7.62)
110 Chapter 7
Chapter 8
Kolmogorov equations
8.1 The deterministic case
We consider here the problem
_
_
_
X
(t) = b(t, X(t)), t ∈ [s, T],
X(s) = x ∈ R
n
,
(8.1)
where s ∈ [0, T) and b : [0, T] R
n
→R
n
fulﬁlls the following hypothesis.
Hypothesis 8.1
(i) b is continuous on [0, T] R
n
.
(ii) There exists M > 0 such that
[b(t, x) −b(t, y)[ ≤ M[x −y[, x, y ∈ R
n
, t ∈ [0, T].
(iii) b is diﬀerentiable with respect to x and b
x
is continuous on [0, T] R
n
.
As well known, under Hypothesis 8.1 problem (8.1) has a unique solution
X() = X(, s, x) ∈ C
1
([s, T]; R
n
), and it holds
X(t, s, x) = X(t, u, X(u, s, x)), 0 ≤ s ≤ u ≤ t ≤ T, x ∈ R
n
. (8.2)
Morever, diﬀerentiating (8.2) with respect to u and setting u = s we ﬁnd
X
s
(t, s, x) + X
x
(t, s, x) b(s, x) = 0, 0 ≤ s ≤ t ≤ T, x ∈ R
n
. (8.3)
Of great interest for the applications is the transition evolution operator
P
s,t
, s, t ∈ [0, T], deﬁned on the space C
b
(R
n
) by
P
s,t
ϕ(x) = ϕ(X(t, s, x)), x ∈ R
n
, s, t ∈ [0, T]. (8.4)
111
112 Kolmogorov equations
As easily checked, P
s,t
is a linear bounded operator on C
b
(R
n
). Moreover for
any ϕ ∈ C
b
(R
n
) the mapping
[0, T] [0, T] R
n
→R
n
, (s, t, x) → P
s,t
ϕ(x),
is continuous. From (8.2) it follows immediately the cocycle property
P
s,t
= P
s,u
P
u,t
, s, t, u ∈ [0, T]. (8.5)
Proposition 8.1 For any ϕ ∈ C
1
b
(R
n
) we have
d
dt
P
s,t
ϕ = P
s,t
L(t)ϕ, t ≥ s (8.6)
and
d
ds
P
s,t
ϕ = −L(s)P
s,t
ϕ, t ≥ s, (8.7)
where
L(t)ϕ(x) = ¸b(t, x), ϕ
x
(x)), ϕ ∈ C
1
b
(R
n
), x ∈ R
n
. (8.8)
Proof. We have
d
dt
P
s,t
ϕ(x) =
d
dt
ϕ(X(t, s, x)) = ¸b(t, X(t, s, x)), ϕ
x
(X(t, s, x)))
and
P
s,t
L(t)ϕ(x) = ¸b(t, X(t, s, x)), ϕ
x
(X(t, s, x))),
so that (8.6) follows.
Let us prove (8.7). We have, taking into acccount (8.3),
d
ds
P
s,t
ϕ(x) =
d
ds
ϕ(X(t, s, x)) = −¸ϕ
x
(X(t, s, x)), X
x
(t, s, x) b(s, x))
= −L(s)P
s,t
ϕ(x).
Let us now consider the following partial diﬀerential equation called trans
port equation
_
_
_
z
s
(s, x) +¸b(s, x), z
x
(s, x)) = 0, s ∈ [0, T]
z(T, x) = ϕ(x),
(8.9)
where ϕ ∈ C
1
b
(R
n
) and T > 0 is ﬁxed.
Chapter 8 113
Theorem 8.2 Assume that b : [0, T] R
n
→ R
n
fulﬁlls Hypothesis 8.1 and
let ϕ ∈ C
1
b
(R
n
). Then problem (8.9) has a unique solution z. z is given by
z(s, x) = P
s,T
ϕ(x) = ϕ(X(T, s, x)), s ∈ [0, T], x ∈ R
n
. (8.10)
Proof Existence. It is enough to notice that z, given by (8.10), is a
solution of (8.9) by (8.6).
Uniqueness. If z is a solution of problem (8.9) we have
d
ds
z(s, X(s, u, x))
= z
t
(s, X(s, u, x)) +¸z
x
(s, X(s, u, x)), X
t
(s, u, x))
= z
t
(s, X(s, u, x)) +¸z
x
(s, X(s, u, x)), b(s, X(s, u, x))) = 0.
Therefore z(s, X(s, u, x)) is constant in s. Setting s = T and s = u we ﬁnd
that z(T, X(T, u, x)) = z(u, X(u, u, x)) which implies z(u, x) = ϕ(X(T, s, x))
as required.
8.1.1 The autonomous case
We assume here that b(t, x) = b(x) and consider the problem
_
_
_
X
(t) = b(X(t)), t ≥ 0,
X(0) = x ∈ R
n
,
(8.11)
whose solution we denote by X(, x). In this case it is easy to check that for
any t > s ≥ 0, we have P
s,t
= P
0,t−s
.
Deﬁne
P
t
ϕ(x) = ϕ(X(t, x)), ϕ ∈ C
b
(R
n
), t ≥ 0, x ∈ R
n
, (8.12)
so that by (8.5) it follows the semigroup law
P
t+s
= P
t
P
s
, t, s ≥ 0. (8.13)
P
t
is called the transition semigroup associated with (8.11). By Proposition
8.1 we deduce
Proposition 8.3 For any ϕ ∈ C
1
b
(R
n
) we have
D
t
P
t
ϕ = P
t
Lϕ = LP
t
ϕ, t ≥ 0 (8.14)
where
Lϕ(x) = ¸b(x), ϕ
x
(x)), ϕ ∈ C
1
b
(R
n
), x ∈ R
n
. (8.15)
114 Kolmogorov equations
Finally, by Theorem 8.2 we have
Theorem 8.4 Assume that b ∈ C
1
b
(R
n
) and let ϕ ∈ C
1
b
(R). Then problem
_
_
_
u
t
(t, x) = ¸b(x), u
x
(t, x)), t ≥ 0, x ∈ R
n
u(0, x) = ϕ(x), x ∈ R
n
.
(8.16)
has a unique solution given by
u(t, x) = P
t
ϕ(x) = ϕ(X(t, x)), t ≥ 0, x ∈ R
n
. (8.17)
8.2 Stochastic case
We consider the stochastic evolution equation
_
_
_
dX(t) = b(t, X(t))dt + σ(t, X(t))dB(t)
X(s) = x ∈ R
d
(8.18)
and assume that the following hypothesis holds.
Hypothesis 8.2 (i) b : [0, T] R
n
→R
n
and σ : [0, T] R
n
→ L(R
r
, R
n
)
are continuous.
(ii) There exists M > 0 such that
[b(t, x)−b(t, y)[+σ(t, x)−σ(t, y)
HS
≤ M[x−y[, x, y ∈ R
n
, t ∈ [0, T].
(iii) b and σ have ﬁrst and second partial derivatives with respect to x con
tinuous and bounded in [0, T] R
n
.
We denote as before by X(, s, x) the solution of (8.18) corresponding to
η = x ∈ R
n
. For all t, s with 0 ≤ s ≤ t ≤ T and for all function ϕ ∈ C
b
(R
n
)
we set
P
s,t
ϕ(x) = E[ϕ(X(t, s, x))], x ∈ R
n
, 0 ≤ s ≤ t ≤ T. (8.19)
As easily checked, P
s,t
is a linear bounded operator on C
b
(R
n
).
P
s,t
, 0 < s ≤ t ≤ T, is called the transition evolution operator associated with
(8.18). By Chapter 6 we know that the mapping
(s, t, x) → P
s,t
ϕ(x),
is continuous for all ϕ ∈ C
b
(R
n
).
Chapter 8 115
8.3 Basic properties of transition operators
Let us introduce the Kolmogorov operator
(L(s)ϕ)(x) =
1
2
Tr [ϕ
xx
(x)σ(s, x)σ
∗
(s, x)] +¸b(s, x), ϕ
x
(x)), ϕ ∈ C
2
b
(R
n
).
(8.20)
The ﬁrst basic identity is the following.
Proposition 8.5 Assume that Hypothesis 8.2 holds and let ϕ ∈ C
2
b
(R
n
).
Then P
s,t
ϕ is diﬀerentiable in t and we have
d
dt
P
s,t
ϕ = P
s,t
L(t)ϕ, t ≥ 0. (8.21)
Proof. By the Itˆ o formula we have that
d
t
ϕ(X(t, s, x)) = (L(t)ϕ)(X(t, s, x)) +¸ϕ
x
(X(t, s, x)), σ(t, X(t, s, x))dB(t)).
Integrating with respect to t and taking expectation, yields
E[ϕ(X(t, s, x))] = ϕ(x) +
_
t
s
E[(L(r)ϕ)(X(r, s, x))]dr,
that is
P
s,t
ϕ(x) = ϕ(x) +
_
t
s
P
r,t
(L(r)ϕ)(x)dr,
which coincides with (8.21).
The second basic identity is the following,
Proposition 8.6 Assume that Hypothesis 8.2 holds and let ϕ ∈ C
2
b
(R
n
).
Then P
s,t
ϕ is diﬀerentiable in s and we have
d
ds
P
s,t
ϕ = −L(s)P
s,t
ϕ, t ≥ 0. (8.22)
Proof. Taking expectation in the backward Itˆ o formula (7.62) we ﬁnd
P
s,t
ϕ(x) −ϕ(x) =
_
t
s
L(r)P
s,r
ϕ(x)dr,
which yields (8.22).
116 Kolmogorov equations
8.4 Parabolic equations
We consider here the parabolic equation
_
_
_
z
s
(s, x) + (L(s)(z(s, )))(x) = 0, 0 ≤ s < T,
z(T, x) = ϕ(x), x ∈ R
n
,
(8.23)
We say that a function z : [0, T] R
n
→ R is a solution to (8.23) if z is
continuous and bounded together with its partial derivatives z
t
, z
x
, z
xx
, and
fulﬁlls (8.23).
Theorem 8.7 Assume that Hypothesis 8.2 holds and let ϕ ∈ C
2
b
(R
n
). Then
there exists a unique solution z of problem (8.23). z is given by
z(s, x) = E[ϕ(X(T, s, x))], 0 < s ≤ T, ϕ ∈ C
2
b
(R
n
). (8.24)
Proof. Existence. By (8.22) it follows that
z(s, x) = P
s,T
ϕ(x), s ∈ [0, T], x ∈ R
n
,
fulﬁlls (8.23).
Uniqueness. Let z be a solution to (8.23), and let 0 ≤ u ≤ s ≤ T. Let us
compute the Itˆ o diﬀerential of z(s, X(s, u, x)). We have
d
s
z(s, X(s, u, x)) = z
s
(s, X(s, u, x))ds + (L(s)z(s, X(s, u, )))(x)
+¸z
x
(s, X(s, u, x)), σ(s, X(s, u, x))dB(s))
= ¸z
x
(s, X(s, u, x)), σ(s, X(s, u, x))dB(s)).
since z fulﬁlls (8.23). Integrating in s between u and T yields
z(T, X(T, u, x)) −z(u, X(u, u, x)) = ϕ(X(t, u, x)) −z(u, x)
=
_
t
u
z
x
(s, X(s, u, x))σ(s, X(s, u, x))dB(s).
Now, taking expectation we ﬁnd
z(u, x) = E[ϕ(X(t, u, x))].
Exercise 8.8 Prove the cocycle law
P
s,r
P
r,t
= P
s,t
(8.25)
for 0 ≤ s ≤ t ≤ t ≤ T.
Chapter 8 117
8.4.1 Autonomous case
Assume that b and σ are independent of t :
b(t, x) = b(x), σ(t, x) = σ(x), x ∈ R
n
.
Then we have L(s) = L where
Lϕ(x) =
1
2
Tr [ϕ
xx
(x)σ(x)σ
∗
(x)] +¸b(x), ϕ
x
(x)), ϕ ∈ C
2
b
(R
n
).
Proposition 8.9 Let X(t, s, x) be the solution of the stochastic evolution
equation
_
_
_
dX(t) = b(X(t))dt + σ(X(t))dB(t)
X(s) = x ∈ R
n
.
(8.26)
Then for any and a > 0 the laws of X(t, s, x) and X(t +a, s +a, x) coincide.
Proof. Set Y (t) = X(t + a, s + a, x). The we have
X(t+a, s+a, x) = x+
_
t+a
s+a
b(X(r, s+a, x))dr+
_
t+a
s+a
σ(X(r, s+a, x))dB(r).
Setting r −a = ρ yields
Y (t) = x +
_
t
s
b(Y (ρ))dρ +
_
t
s
σ(Y (ρ))d[B(ρ + a) −B(a)].
Setting B
1
(t) = B(t + a) − B(a) we see that Y (t) fulﬁlls equation (8.26)
but with the Brownian motion B(t) replaced by B
1
(t). Now the conclusion
follows.
By the proposition and the cocycle law (8.25)it follows that, setting
P
t
= P
0,t
, t ≥ 0,
we have
P
t+s
= P
t
P
s
, t, s ≥ 0, P
0
= 1.
Thus P
t
, t ≥ 0 is a semgroup of linear operators in C
b
(R
d
).
Setting
v(s, x) = u(t, t −s, x), t ≥ 0, s ∈ [0, t], x ∈ R
n
,
problem (8.23) becomes
_
_
_
v
s
(s, x) = Lv(s, x), s ∈ [0, t], x ∈ R
n
,
v(0, x) = ϕ(x), x ∈ R
(8.27)
Then by Theorem 8.7 we ﬁnd the result
118 Kolmogorov equations
Theorem 8.10 Assume that b, σ : R → R are Lipschitz continuous and of
class C
2
. Then, for any ϕ ∈ C
2
b
(R), problem (8.27) has a unique solution
given by
v(s, x) = P
t−s,t
ϕ(x) = P
t
ϕ(x), t ≥ 0, s ∈ [0, t], x ∈ R. (8.28)
8.5 Examples
Example 8.11 Consider the parabolic equation in R
n
_
_
_
u
t
(t, x) =
1
2
Tr [Qu
xx
(t, x)] +¸Ax + u
x
(t, x))
u(0, x) = ϕ(x),
(8.29)
where A, Q ∈ L(R
n
), Q is symmetric and ¸Qx, x) ≥ 0 for all x ∈ R
n
.
The corresponding stochastic diﬀerential equation is
_
_
_
dX(t) = AX(t)dt +
√
Q dB(t),
X(0) = x,
(8.30)
where B is a standard Brownian motion in a probability space (Ω, G, P)
taking values in R
n
. The solution of (8.30) is given by the variation of
constants formula
X(t, x) = e
tA
x +
_
t
0
e
(t−s)A
_
QdB(s). (8.31)
Therefore the law of X(t, x) is given by
X(t, x)
#
P = N
e
tA
x,Q
t
, (8.32)
where
Q
t
=
_
t
0
e
sA
Qe
sA
∗
ds, t ≥ 0, (8.33)
where A
∗
is the adjoint of A.
Consequently, the transition semigroup P
t
looks like
P
t
ϕ(x) =
_
R
n
ϕ(y)N
e
tA
x,Q
t
(dy). (8.34)
So, the solution of (8.29) is given by
u(t, x) = P
t
ϕ(x).
Chapter 8 119
If, in particular, det Q
t
> 0 we have
u(t, x) = (2π)
−n/2
[det Q
t
]
−1/2
_
R
n
e
−
1
2
Q
−1
t
(y−e
tA
x),(y−e
tA
x)
ϕ(y)dy. (8.35)
Example 8.12 Consider the parabolic equation in R
_
_
_
u
t
(t, x) =
1
2
qx
2
u
xx
(t, x) + axu
x
(t, x)
u(0, x) = ϕ(x),
(8.36)
where q > 0 and a ∈ R.
The corresponding stochastic diﬀerential equation is
_
_
_
dX(t) = aX(t)dt +
√
q X(t)dB(t),
X(0) = x,
(8.37)
where B is a real Brownian motion in is a real Brownian motion in some
probability space (Ω, F, P).
The solution of (8.37) is given by
X(t, x) = e
(a−q/2)t+
√
q B(t)
x. (8.38)
Therefore
P
t
ϕ(x) =
1
√
2πt
_
+∞
−∞
e
−
y
2
2t
ϕ(e
(a−q/2)t+
√
q y
x)dy. (8.39)
120 Kolmogorov equations
Appendix A
λsystems and πsystems
Let Ω be a non empty set. A non empty family R of parts of Ω is called a
πsystem if
A, B ∈ R =⇒ A ∩ B ∈ R,
a λsystem if
_
¸
¸
¸
¸
_
¸
¸
¸
¸
_
(i) Ω, ∅ ∈ D.
(ii) A ∈ D =⇒ A
c
∈ D.
(iii) (A
i
) ⊂ D mutually disjoint =⇒
∞
i=1
A
i
∈ D.
(A.1)
Obviously any algebra is a πsystem. Moreover, if D is a λsystem such that
A, B ∈ D =⇒ A ∩ B ∈ D then it is σ–algebra. In fact if (A
i
) is a sequence
in D of not necessarily disjoint sets we have
∞
_
i=1
A
i
= A
1
∪ (A
2
¸ A
1
) ∪ (A
3
¸ A
2
¸ A
1
) ∪ ∈ D
and so
∞
i=1
A
i
∈ D by (ii) and (iii).
Let us prove the following Dynkin theorem.
Theorem A.1 Let R be a πsystem and let D be a λsystem including R.
Then we have σ(R) ⊂ D, where σ(R) is the σ algebra generated by R. If in
particular, D ⊂ σ(R) we have σ(R) = D.
Proof. Let D
0
be the minimal λsystem including R. We are going to show
that D
0
is a σ–algebra, which will imply the theorem. For this it is enough
to show, as remarked before, that the following inclusion holds
A, B ∈ D
0
=⇒ A ∩ B ∈ D
0
. (A.2)
121
122 λsystems and πsystems
For any B ∈ D
0
we set
H (B) = ¦F ∈ D
0
: B ∩ F ∈ D
0
¦.
We claim that H (B) is a λsystem. In fact properties (i) and (iii) are clear.
It remains to show that if F ∩ B ∈ D
0
then F
c
∩ B ∈ D
0
or, equivalently,
that F ∪B
c
∈ D
0
. In fact, since F ∪B
c
= (F ¸ B
c
) ∪B
c
= (F ∩B) ∪B
c
and
F ∩ B and B
c
are disjoint, we have that F ∪ B
c
∈ D
0
as required.
If we show that
H (B) ⊃ R, ∀B ∈ D
0
(A.3)
then we conclude that H (B) = D
0
by the minimality of D
0
and (A.2) is
proved.
On the other hand it is clear that if R ∈ R we have R ⊂ H (R) since R is
a πsystem. Therefore H (R) = D
0
by the minimality of D
0
. Consequently,
the following implication holds
R ∈ R, B ∈ D
0
⇒ R ∩ B ∈ D
0
,
which yields R ⊂ H (B) and (A.3) is fulﬁlled.
Example A.2 Let A be an algebra of subsets of Ω and let F be the σ
algebra generated by A. Let P
1
and P
2
be probability measures on (Ω, F)
such that
P
1
(I) = P
2
(I), ∀ I ∈ A.
Using the Dynkin theorem we can show that P
1
= P
2
. It is clear in fact that
A is a πsystem. Deﬁne
D = ¦B ∈ F : P
1
(B) = P
2
(B)¦.
It is easy to see that D is a λsystem which contains D. So, by Corollary
A.1 it follows that P
1
= P
2
.
Appendix B
Conditional expectation
B.1 Deﬁnition
We are given a probability space (Ω, F, P) and a σalgebra G included in F.
Let X : Ω →R be a real random variable on (Ω, F, P)
(1)
.
We say that X is Gmeasurable if
I ∈ B(R) ⇒ X
−1
(I) ∈ F.
It is clear that X is not Gmeasurable in general.
Let us consider the signed measure
µ(G) =
_
G
XdP, G ∈ G.
It is clear that µ is absolutely continuous with respect to the restriction of
P to G. Therefore, by the RadonNikodym Theorem there exists a unique
Y ∈ L
1
(Ω, G, P) such that
µ(G) =
_
G
XdP =
_
G
Y dP, ∀ G ∈ G. (B.1)
The Gmeasurable random variable Y is called the conditional expectation of
X given G; it is denoted by E(X[G).
In view of (B.1) E(X[G) is characterized by
_
G
XdP =
_
G
E(X[G)dP, ∀ G ∈ G. (B.2)
Exercise B.1 Assume that X ∈ L
2
(Ω, F, P). Show that E(X[G) coincides
with the orthogonal projection of X into the closed subspace L
2
(Ω, G, P) of
L
2
(Ω, F, P).
(1)
In all this appendix by random variable we mean an equivalence class of random
variables with respect to the usual equivalence relation.
123
124 Conditional expectation
B.2 Basic properties
Let X, Y ∈ L
1
(Ω, F, P) and let G be σalgebra included in F. It is obvious
that if X is Gmeasurable, we have E(X[G) = X. Setting G = Ω in (B.2)
yields
E[E(X[G)] = E(X). (B.3)
Moreover, one can check easily the linearity of conditional expectation,
E(αX + βY [G) = αE(X[G) + βE(Y [G), (B.4)
for all α, β ∈ R and all X, Y ∈ L
1
(Ω, F, P). Also if X ≥ 0, Pa.s., one has
E(X[G) ≥ 0, Pa.s. From this one deduces the inequality
[E(X[G)[ ≤ E([X[ [G). (B.5)
Proposition B.2 Assume that X is independent of G. Then we have
E(X[G) = E(X). (B.6)
Proof. Let A ∈ G. Then 1l
A
and X are independent so that
_
A
XdP =
_
Ω
1l
A
XdP = P(A)E(X) =
_
A
E(X[G)dP.
Proposition B.3 Let H be a σalgebra included in G. Then we have
E(X[H ) = E
_
E(X[G)
¸
¸
H
¸
. (B.7)
Proof. Let A ∈ H . Then we have
_
A
XdP =
_
A
E(X[H )dP (B.8)
and
_
A
XdP =
_
A
E(X[G)dP =
_
A
E
_
E(X[G)
¸
¸
H
¸
dP. (B.9)
So, comparing (B.8) and (B.9) we see that
_
A
E(X[H )dP =
_
A
XdP =
_
A
E
_
E(X[G)
¸
¸
H
¸
dP.
Proposition B.4 Let X, Y, XY ∈ L
1
(Ω, F, P). Assume that X is Gmeasurable.
Then we have
E(XY [G) = XE(Y [G). (B.10)
Appendix B 125
Proof. It is enough to show (B.10) for X = 1l
A
where A ∈ G. Let now
G ∈ G, then since G∩ A ∈ G we have
_
G
E(1l
A
Y [G)dP =
_
G
1l
A
Y dP =
_
G∩A
Y dP
=
_
G∩A
E(Y [G)dP =
_
G
1l
A
E(Y [G)dP,
for any G ∈ G.
Recalling Proposition B.2 we ﬁnd.
Corollary B.5 Let X, Y, XY ∈ L
1
(Ω, F, P). Assume that X is Gmeasurable
and that Y is independent of G. Then we have
E(XY [G) = XE(Y ). (B.11)
Let us prove now a useful generalization of this Corollary.
Proposition B.6 Let X, Y ∈ L
1
(Ω, F, P) and let φ : R
2
→ R be bounded
and Borel. Assume that X is Gmeasurable and Y is independent of G. Then
we have
E(φ(X, Y )[G) = h(X), (B.12)
where
h(x) = E[φ(x, Y )], x ∈ R. (B.13)
Proof. We have to show that
_
G
φ(X, Y )dP =
_
G
h(X)dP, ∀ G ∈ G.
This is clearly equivalent to
E(Zφ(X, Y )) = E(Zh(X)), ∀Z ∈ L
1
(Ω, G, P). (B.14)
Denote by µ the law of the random variable (X, Y, Z) with values in R
3
µ = (X, Y, Z)
#
P.
So,
E(Zφ(X, Y )) =
_
R
3
zφ(x, y)µ(dx, dy, dz). (B.15)
126 Conditional expectation
Since X and Z are Gmeasurable and Y is independent of G, the random
variables (X, Z) and Y are independent so that
µ(dx, dy, dz) = ν(dx, dz)λ(dy),
where
ν(dx, dz) = (X, Z)
#
P(dx, dz), λ(dy) = Y
#
P(dy).
Therefore we can write (B.15) as
E(Zφ(X, Y )) =
_
R
3
zφ(x, y)ν(dx, dz)λ(dy).
Using the Fubini Theorem we get ﬁnally
E(Zφ(X, Y )) =
_
R
2
z
__
R
φ(x, y)λ(dy)
_
ν(dx, dz)
=
_
R
2
zh(x)ν(dx, dz) = E(Zh(X)),
as required.
Exercise B.7 Let F, H, FH ∈ L
1
(Ω, G, P) and Z = E(H[G). Prove that
E(FH) = E(FZ). (B.16)
Exercise B.8 Let g : R → R be convex and let F, g(F) ∈ L
1
(Ω, F, P).
Prove the Jensen inequality
E(g(F)[G) ≥ g(E(F[G)). (B.17)
Appendix C
Martingales
C.1 Deﬁnitions
Let (Ω, F, P) be a probability space, (F
t
)
t≥0
an increasing family of σ
algebras included in F and (M(t))
t∈[0,T]
with M(t) ∈ L
1
(Ω, F
t
, P), t ∈ [0, T],
a stochastic process.
(M(t))
t∈[0,T]
is said to be a martingale (with respect to the ﬁltration
(F
t
)
t≥0
) if
E[M(t)[F
s
] = M(s), ∀ 0 ≤ s < t ≤ T,
a submartingale if
E[M(t)[F
s
] ≥ M(s), ∀ 0 ≤ s < t ≤ T,
a supermartingale if
E[M(t)[F
s
] ≤ M(s), ∀ 0 ≤ s < t ≤ T.
Thus (M(t))
t∈[0,T]
is a martingale if and only if
_
A
M(s)dP =
_
A
M(t)dP, ∀ 0 ≤ s < t ≤ T, A ∈ F
s
,
a submartingale if and only if
_
A
M(s)dP ≥
_
A
M(t)dP, ∀ 0 ≤ s < t ≤ T, A ∈ F
s
,
and a supermartingale if and only if
_
A
M(s)dP ≤
_
A
M(t)dP, ∀ 0 ≤ s < t ≤ T, A ∈ F
s
.
127
128 Martingales
Proposition C.1 If M is a martingale then [M[ is a submartingale.
Proof. Let 0 ≤ s < t ≤ T, A ∈ F
s
. Set
A
+
= ¦ω ∈ Ω : M(s)(ω) > 0¦, A
−
= ¦ω ∈ Ω : M(s)(ω) ≤ 0¦.
Clearly A
+
and A
−
belong to F
s
. Consequently we have
_
A
[M(s)[dP =
_
A
+
M(s)dP −
_
A
−
M(s)dP
=
_
A
+
M(t)dP −
_
A
−
M(t)dP ≤
_
A
[M(t)[dP.
This shows that [M[ is a submartingale.
Example C.2 The Brownian motion B is a martingale. In fact, let t > s
and A ∈ F
s
. Since B(t) −B(s) and 1l
A
are independent we have
_
A
(B(t) −B(s))dP = E(1l
A
(B(t) −B(s))) = 0,
so that
_
A
B(t)dP =
_
A
B(s)dP.
Exercise C.3 Using Jensen’s inequality prove that any convex function of
a martingale is a submartingale. (See Exercise B.8).
C.2 The basic inequality for martingales
Let M(t) be a martingale, let 0 < t
1
< t
2
< ... < t
n
≤ T and set
S = sup
1≤i≤n
[M(t
i
)[.
We are going to prove an important estimate (due to Kolmogorov) of S in
terms of M(t
n
).
Proposition C.4 For all λ > 0 we have
P(S ≥ λ) ≤
1
λ
_
{S≥λ}
[M(t
n
)[dP. (C.1)
Appendix C 129
Proof. Set
A
1
= ¦[M(t
1
)[ ≥ λ¦,
A
2
= ¦[M(t
1
)[ < λ, [M(t
2
)[ ≥ λ¦,
A
n
= ¦[M(t
1
)[ < λ, ..., [M(t
n
)[ ≥ λ¦.
Clearly, sets A
1
, ..., A
n
are mutually disjoint. Moreover A
i
∈ F
t
i
, i = 1, ..., n,
and we have
¦S ≥ λ¦ =
n
_
i=1
A
i
.
Let us estimate
_
{S≥λ}
[M(t
n
)[dP. We have obviously
_
A
n
[M(t
n
)[dP ≥ λP(A
n
).
Now we estimate
_
A
n−1
[X(t
n
)[dP. We have, recalling that [M(t)[ is a sub–
martingale,
λP(A
n−1
) ≤
_
A
n−1
[M(t
n−1
)[dP ≤
_
A
n−1
[M(t
n
)[dP.
Therefore
_
A
n−1
[M(t
n
)[dP ≥ λP(A
n−1
).
Proceeding in a similar way we obtain
_
A
k
[M(t
n
)[dP ≥ λP(A
k
), k = 1, . . . , n. (C.2)
Summing up on k from 1 to n the conclusion follows.
C.3 Square integrable martingales
In this section we are given a martingale M(t) such that M(t) ∈ L
2
(Ω, F, P)
for all t ∈ [0, T].
Let 0 < t
1
< t
2
< ... < t
n
≤ T and set as before
S = sup
1≤i≤n
[M(t
i
)[.
We are going to estimate of E[S
2
] in terms of E[M
2
(t
n
)].
130 Martingales
Proposition C.5 We have
E
_
sup
1≤i≤n
[M(t
i
)[
2
_
≤ 4E([M(t
n
)[
2
). (C.3)
Proof. Set
F(t) = P(S > t), t ≥ 0.
By (C.1) we have
F(t) ≤
1
t
_
{S≥t}
[M(t
n
)[dP. (C.4)
Consequently
E(S
2
) =
_
∞
0
P(S
2
> t)dt =
_
∞
0
P(S >
√
t)dt.
So, by (C.1) and the Fubini Theorem we have
E(S
2
) ≤
_
∞
0
_
1
√
t
_
{S≥
√
t}
[M(t
n
)[dP
_
dt
=
_
[0,+∞)×Ω
1
√
t
[M(t
n
)[1l
{S≥
√
t}
P(dω)dt
=
_
Ω
[M(t
n
)[P(dω)
_
∞
0
1
√
t
1l
{S≥
√
t}
dt
=
_
Ω
[M(t
n
)[P(dω)
_
S
2
0
1
√
t
dt
= 2
_
Ω
[M(t
n
)SP(dω) ≤ 2
__
Ω
[M(t
n
)[
2
dP
_
1/2
__
Ω
S
2
dP
_
1/2
.
Now the conclusion follows easily.
Corollary C.6 Let M be a square integrable continuous martingale. Then
for any T > 0 we have
E
_
sup
t∈[0,T]
[M(t)[
2
_
≤ 4E[M
2
(T)]. (C.5)
Appendix C 131
Proof. Let 0 < s
1
< s
2
< < s
m
= T. By Proposition C.5 it follows that
E
_
sup
1≤i≤m
[M(s
i
)[
2
_
≤ 4E
_
[M(T)[
2
¸
.
Since M is continuous it follows, by the arbitrariness of the sequence s
1
, s
2
, . . . , s
m
,
that
E
_
sup
s∈[0,T]
[M(s)[
2
_
≤ 4E
_
[M(T)[
2
¸
,
as required.
132 Martingales
Appendix D
Fixed points depending on
parameters
D.1 Introduction
Let Λ, E be Banach spaces (norms [ [). We are given a continuous mapping
F : Λ E → E, (λ, x) → F(λ, x)
and assume that
Hypothesis D.1 There exists κ ∈ [0, 1) such that
[F(λ, x) −F(λ, y)[ ≤ κ[x −y[, ∀ λ ∈ Λ, x, y ∈ E.
The following result (contraction principle) is classical.
Theorem D.1 (i). There exists a unique continuous mapping
x : Λ → E, λ → x(λ),
such that
x(λ) = F(λ, x(λ)), ∀ λ ∈ Λ. (D.1)
(ii). If in addition F is of class C
1
, then x is of class C
1
and
x
(λ) = F
λ
(λ, x(λ)) + F
x
(λ, x(λ))x
(λ). (D.2)
We want to generalize the second part of this result to mappings F(λ, x)
which are only continuously Gˆateaux diﬀerentiable.
133
134 Fixed points
D.2 Gˆateaux diﬀerentiable mappings
Let A and B be Banach spaces and let Φ : A → B be a continuous mapping
from A into B.
Deﬁnition D.2 We say that Φ is Gˆ ateaux diﬀerentiable if there exists a
mapping
DΦ : A → L(A, B), a → DΦ(a),
such that
lim
ξ→0
1
ξ
(Φ(a + ξc) −Φ(a)) = DΦ(a)c, ∀ a, c ∈ A.
If in addition for all c ∈ A the mapping A → B, a → DΦ(a)c is continuous
we say that Φ is continuously Gˆ ateaux diﬀerentiable.
Remark D.3 It is well known that if the mapping A → L(A, B), a →
DΦ(a) is continuous then Φ is diﬀerentiable.
(1)
Example D.4 Let A, B = L
2
(0, 1) and Φ(x) = sin x. Then one can check
easily that Φ is continuously Gˆ ateaux diﬀerentiable and
DΦ(x)y = y cos x, ∀ x, y ∈ L
2
(0, 1).
However, (as one can see) Φ is not diﬀerentiable in any point.
We shall need the following result.
Proposition D.5 Let Φ : A → B be continuously Gˆateaux diﬀerentiable.
Then the following identity holds
Φ(c) −Φ(a) =
_
1
0
DΦ((1 −ξ)a + ξc)(c −a)dξ. (D.3)
Proof. Set
F(ξ) = Φ((1 −ξ)a + ξc), ξ ∈ [0, 1].
Then we have
F
(ξ) = DΦ((1 −ξ)a + ξc)(c −a)dξ,
and the conclusion follows just integrating this identity between 0 and 1.
(1)
One also says that Φ is Fr´echet diﬀerentiable.
Appendix D 135
D.3 The main result
We can back to the notations of the introduction and consider two Banach
spaces Λ and E and a continuous mapping
F : Λ E → E, (λ, x) → F(λ, x).
We assume that Hypothesis D.1 is fulﬁlled and denote by x the mapping
x : Λ → E, λ → x(λ),
such that
x(λ) = F(λ, x(λ)), ∀ λ ∈ Λ. (D.4)
Theorem D.6 Assume that Hypotheses D.1 is fulﬁlled and that F is con
tinuously Gˆateaux diﬀerentiable. Then x() is continuously Gˆateaux diﬀeren
tiable as well and we have
x
(λ) µ = (1 −F
x
(λ, x(λ)))
−1
F
λ
(λ, x(λ)) µ, (D.5)
equivalently
x
(λ) µ = F
λ
(λ, x(λ)) µ + F
x
(λ, x(λ))(x
(λ) µ). (D.6)
Proof. Let λ, µ ∈ Λ and h ∈ R. From (D.4) and (D.3) it follows that
x(λ + hµ) −x(λ) = F(λ + hµ, x(λ + hµ)) −F(λ, x(λ))
= h
_
1
0
F
λ
(λ + ξhµ, x(λ) + ξ(x(λ + hµ) −x(λ))) µdξ
+
_
1
0
F
x
(λ + ξhµ, x(λ) + ξ(x(λ + hµ) −x(λ))) (x(λ + hµ) −x(λ))dξ.
(D.7)
Set now
G(λ, x, µ, h)z = Gz :=
_
1
0
F
x
(λ+ξhµ, x(λ)+ξ(x(λ+hµ)−x(λ)))zdξ, z ∈ E.
Then G ∈ L(E) and by Hypothesis D.1
[Gz[ ≤ κ[z[, ∀ z ∈ E.
136 Fixed points
Then from equation (D.7) we have
(1 −G(λ, x, µ, h))(x(λ + hµ) −x(λ))
= h
_
1
0
F
λ
(λ + ξhµ, x(λ) + ξ(x(λ + hµ) −x(λ))) µdξ,
which implies
1
h
x(λ + hµ) −x(λ)) = (1 −G(λ, x, µ, h))
−1
_
1
0
F
λ
(λ + ξhµ, x(λ) + ξ(x(λ + hµ) −x(λ))) µdξ.
Letting h → 0 we ﬁnd
x
(λ) µ = (1 −F
x
(λ, x(λ)))
−1
F
λ
(λ, x(λ)).
Therefore
x
(λ) µ −F
x
(λ, x(λ))(x
(λ) µ) = F
λ
(λ, x(λ)).
Appendix E
Fractional Sobolev spaces and
regularity of processes
E.1 Fractional Sobolev spaces on [0, 1]
Let ∈ (0, 1), m ∈ N. Deﬁne
f
2m
,2m
:=
_
[0,T]
2
[f(t) −f(s)[
2m
[t −s[
1+2m
dt ds
W
,2m
(0, T) is by deﬁnition the space of all f : [0, T] →R such that f
,2m
<
+∞.
Theorem E.1 (Sobolev embedding) Assume that > 1/(2m). Then the
following inclusion holds with continuous embedding.
W
,2m
(0, T) ⊂ C
−1/(2m)
([0, T]). (E.1)
Example E.2 (The Brownian motion) Let > 0 and let p ≥ 1. We ask
the question whether B() belongs to W
,p
(0, T) or not.
Let us compute
E(B
p
W
,p) = E
_
[0,T]
2
[B(t) −B(s)[
p
[t −s[
1+p
dt ds
Take for simplicity p = 2m, then
E
_
B
2m
W
,2m
_
= E
_
[0,T]
2
[B(t) −B(s)[
2m
[t −s[
1+2m
dt ds
= c
m
_
[0,T]
2
[t −s[
m
[t −s[
1+2m
dt ds = c
m
_
[0,T]
2
[t −s[
m−1−2m
dt ds
137
138 Fractional Sobolev spaces
The integral is ﬁnite if and only if <
1
2
.
For instance taking m = 1 we conclude that B() ∈ W
,2
(0, T) for <
1
2
.
This does not imply that B() is continuous.
But if we take m = 2 we have B() ∈ W
,4
(0, T) again for <
1
2
. Therefore
if
1
4
< <
1
2
we conclude by the Sobolev embedding that B() ∈ C
−
1
4
(0, T).
Arguing similarly taking larger m we conclude that B() ∈ C
α
(0, T) for
any α ∈ (0, 1/2).
E.2 Processes belonging to W
,2m
(0, T)
Let (Ω, F, P) be probability space and let X(t), t ∈ [0, T], be a real stochastic
process on (Ω, F, P). One situation often encountered is when the following
estimate holds for some m > 1, ∈ (0, 1/2), and c
m
> 0
E[[X(t) −X(s)[
2m
] ≤ c
m
[t −s[
m
, ∀ t, s ∈ [0, T]. (E.2)
This estimate (provided m > 1) allows us to conclude that trajectories of X
are H¨older continuous almost surely, as the next proposition shows.
Proposition E.3 Assume that there is m > 1, ∈ (0, 1/2), and c
m
> 0
such that (E.2) is fulﬁlled. Then we have
E
_
[X[
2m
,2m
¸
< +∞. (E.3)
Moreover, X(, ω) belongs to C
−1/(2m)
([0, T]) for almost ω ∈ Ω.
Proof. We have in fact
E
_
X
2m
,2m
_
≤ c
m
_
[0,T]
2
[t −s[
m−1−2m
dt ds < ∞,
since ∈ (0, 1/2) and m− 1 − 2m > −1. The last statement follows from
the Sobolev embedding theorem.
Remark E.4 Kolomogorov test It is a generalization Proposition E.3. As
sume that there is a > 0, b > 0 such that
E[[X(t) −X(s)[
1+a
] ≤ c
m
[t −s[
1+b
∀ t, s ∈ [0, T]. (E.4)
Then X has αH¨ older continuous trajectories with α <
1+b
a
.
Appendix F 139
E.3 Multi dimensional Sobolev spaces and reg
ularity of random ﬁelds
Let ∈ (0, 1), m ∈ N, d ∈ N. Deﬁne
f
2m
,2m
:=
_
[0,T]
2d
[f(x) −f(y)[
2m
[x −y[
d+2m
dx dy.
W
,2m
([0, T]
d
) is by deﬁnition the space of all f : [0, T]
d
→ R such that
f
,2m
< +∞.
Theorem E.5 (Sobolev embedding) Assume that > d/(2m). Then the
following inclusion holds with continuous embedding.
W
,2m
([0, T]
d
) ⊂ C
−d/(2m)
([0, T]
d
). (E.5)
Let (Ω, F, P) be probability space and let X(x), x ∈ [0, T]
d
, be a random
ﬁeld on (Ω, F, P).
Assume that there is m > 1, ∈ (0, 1), and c
m
> 0
E[[X(x) −X(y)[
2m
] ≤ c
m
[t −s[
2m
, ∀ t, s ∈ [0, T]. (E.6)
This estimate implies that almost all trajectories of X are H¨ older continuous
almost surely.
Proposition E.6 Assume that there is m > 1, ∈ (0, 1), and c
m
> 0 such
that (E.2) is fulﬁlled. Then we have
E
_
[X[
2m
,2m
¸
< +∞. (E.7)
Moreover, X(, ω) belongs to C
−d/(2m)
([0, T]) for almost ω ∈ Ω.
Proof. We have in fact
E(X
2m
,2m
) ≤ c
m
_
[0,T]
2
[t −s[
m−1−2m
dt ds < ∞,
since ∈ (0, 1/2) and m− 1 − 2m > −1. The last statement follows from
the Sobolev embedding theorem.
Contents
1 Gaussian measures in Hilbert spaces 1.1 Some concepts of Probability . . . . . . . . . . . 1.1.1 Random variables . . . . . . . . . . . . . . 1.1.2 Product measures . . . . . . . . . . . . . . 1.2 Probability measures in Hilbert spaces . . . . . . 1.2.1 Mean and covariance . . . . . . . . . . . . 1.2.2 Finite dimensional projections of measures 1.3 Gaussian probability measures . . . . . . . . . . . 1.3.1 Gaussian probability measures in R . . . . 1.3.2 Gaussian probability measures in Rn . . . 1.3.3 Gaussian probability measures in H . . . . 1.3.4 Computation of some Gaussian integrals . 1.3.5 The Cameron–Martin space . . . . . . . . 3 3 3 5 5 5 7 9 9 10 11 11 13 17 17 18 18 21 21 22 23 23 25 27 27 28 29 29 31
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
2 Gaussian random variables 2.1 Notations . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Independence . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Independent real variables . . . . . . . . . . . 2.2.2 Independent Gaussian random variables . . . 2.3 Gaussian random variables deﬁned in a Hilbert space 2.3.1 Aﬃne changes of variables . . . . . . . . . . . 2.4 The white noise function . . . . . . . . . . . . . . . . 2.4.1 Equivalence classes of random variables . . . . 2.4.2 Deﬁnition of the white noise function . . . . . 3 Brownian Motion 3.1 Stochastic Processes . . . . . . . . . . . . . . 3.2 Brownian motion . . . . . . . . . . . . . . . . 3.2.1 Construction of a Brownian motion . . 3.2.2 Some properties of a Brownian motion 3.3 Wiener integral . . . . . . . . . . . . . . . . . i
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
ii 3.4 3.5 Continuity of Brownian motion . . . . . . . . . . . . . The standard Brownian motion . . . . . . . . . . . . . 3.5.1 Some properties of C0 . . . . . . . . . . . . . . 3.5.2 The Wiener measure and the standard Brownian Quadratic variation of the Brownian motion . . . . . . Multidimensional Brownian motions . . . . . . . . . . . . . . . . . . . . . . . motion . . . . . . . . 35 36 37 37 39 41 43 43 44 46 49 50 51 52 53 56 57 58 59 61 61 61 63 66 67 70 70 71 72
3.6 3.7
4 Markov property of the Brownian motion 4.1 Filtration . . . . . . . . . . . . . . . . . . . . . . 4.1.1 Ft measurable random variables . . . . . . 4.2 Stopping times . . . . . . . . . . . . . . . . . . . 4.3 The Brownian motion W (t + τ ) − W (τ ) . . . . . 4.4 Transition semigroup . . . . . . . . . . . . . . . . 4.5 Markov property . . . . . . . . . . . . . . . . . . 4.5.1 Strong Markov property . . . . . . . . . . 4.6 Some consequences of the strong Markov property 4.7 Application to partial diﬀerential equations . . . . 4.7.1 The Dirichlet problem in the halfline . . . 4.7.2 The Neumann problem . . . . . . . . . . . 4.7.3 The Ventzell problem . . . . . . . . . . . . 5 The Itˆ integral o 5.1 Deﬁnition of Itˆ’s integral . . . . . . . . . . . . . o 5.1.1 Itˆ’s integral for elementary processes . . . o 5.1.2 General deﬁnition of Itˆ’s integral . . . . . o 5.2 Itˆ integral for mean square continuous processes o 5.3 The Itˆ integral as a stochastic process . . . . . . o 5.4 Itˆ integral with stopping times . . . . . . . . . . o 5.4.1 Stopping times . . . . . . . . . . . . . . . 5.4.2 Itˆ’s integral with stopping times . . . . . o 5.5 Multidimensional Itˆ integrals . . . . . . . . . . . o
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
6 The Itˆ formula o 75 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 6.1.1 The Itˆ formula for unbounded functions . . . . . . . . 82 o 6.2 Itˆ’ formula for a vector valued process . . . . . . . . . . . . . 84 o 7 Stochastic evolution equations 89 7.1 Existence and uniqueness . . . . . . . . . . . . . . . . . . . . . 90 7.1.1 Solution of the stochastic diﬀerential equation in the space CB ([s, T ]; L2m (Ω; Rd )). . . . . . . . . . . . . . . 94
. . . . . . . . . . . . . . . . . .2 7. A λsystems and πsystems . . . . 107 o 111 . . . . . . . . . . . . . .2 Examples . . . . . . . .2 The stochastic case . . . . . . . . . . 118 121 7. . . . . . . . 123 B. . . . . . .1. x) . . . . . . . . . . . . . . . . . . . . . . . . . . .4 7. . . . . . . . . .1 7. . . . . . 113 . . .4. . . . . . . . .1 Deﬁnition . .2. 128 C. . . . . . . . . . . . . . . . 8. 133 . . . . . . . . . . . . . 8. . .1 The deterministic case . 135 E Fractional Sobolev spaces and regularity of processes 137 E. s. 124 C Martingales 127 C. . 101 7. . . . . . . 116 . 1] . .5. . .1 The deterministic case . . . . . . . . . . . . . . . . . . . 8. . . . . . . . . . . . .3 Backward Itˆ’s formula . . . . .3 Square integrable martingales . .2 Gˆteaux diﬀerentiable mappings . . . . . . . .3 Diﬀerential stochastic equations with random coeﬃcients 96 Continuous dependence on data . . . 8. . . . .1 Introduction . . .1. . . . . . . . . . . . . 106 7. . . . . . . . x) . . . . . . . . .5. . .4. . . . 134 . . . . . . . . . . . . . 117 . . . .5.1 Deﬁnitions . . . . . .5 8 Kolmogorov equations 8. . . . . . . . . . . . . . . .4 Parabolic equations . . . . . . . . . . 127 C. . . . . . . 8. . .1 Existence of Xx (t. 105 7. . . . . .3 The main result . . . . . . . .3 Basic properties of transition operators 8. . . 101 7. . . . x) with respect to s. . . . . . . . .2 Processes belonging to W . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Basic properties . . . s.4. . . . . . . 111 . . . . . . . . . . . .1. . . . . . . . . s. . . . . . . . x) with respect to x . . . . . . 114 . . . 105 o 7. . . . 97 7. . . . . .2 The basic inequality for martingales . . . . . . . . . . .5 Examples . . . s. . . . . . . . 133 . .1 Continuous dependence on mean square . 94 7. . .1 Fractional Sobolev spaces on [0. 102 Itˆ Diﬀerentiability of X(t. . . . . . . . . . .3 7. . . . . . . . .1 The autonomous case . . . .2 Stochastic case . . . B Conditional expectation 123 B. . . . . . . . . . . . . . . . . . . . . . 137 E.2m (0. . . . .1 Autonomous case . . . . . . . . a D. . . . . . . . 138 . . . . . . . . . . D. . . . 129 D Fixed points depending on parameters D. . .2 Existence of Xxx (t. . . . 115 . . . T ) . . 97 Almost sure continuity and h¨lderianity of trajectories . 100 o Diﬀerentiability of X(t. . . . . .
3 Multi dimensional Sobolev spaces and regularity of random ﬁelds . . . . . . . . . . . . . . . . . . .2 E. . . . . . . . . . 139 . . . . .
Finally. We recall that T ∈ L(H) is said to be symmetric if T x. The set of all symmetric and positive elements of L(H) will be denoted by L+ (H).Chapter 1 Gaussian measures in Hilbert spaces We shall denote by H a real separable Hilbert space (with inner product ·. endowed with the norm ϕ 0: = sup ϕ(x). y ∈ H. · and norm  · ). x∈H is a Banach space. x=1 T x. positive if T x. 1. 3 . endowed with the norm T = sup x∈H.1. P) be a probabilty space and let E be a Polish (complete separable metric) space. and by L(H) the Banach algebra of all linear bounded operators T : H → H.1 Some concepts of Probability Random variables Let (Ω. we shall denote by Cb (H) the space of all functions ϕ : H → R which are continuous and bounded. Next section is devoted to some basic facts from Measure Theory and Probability needed in what follows. F . we shall denote by B(E) the σ–algebra generated by all closed (or equivalently open) subsets of E. Cb (H).1 1. T y for all x. x ≥ 0 for all x ∈ H. y = x. The elements of B(E) are called Borel sets.
Then we have ϕ(X(ω))P(dω) = Ω E (1) ϕ(x)(X# P)(dx). cn ≥ 0 and I1 . . Let us prove the following basic change of variables formula. P). B(E)) deﬁned as (X# P)(I) = P(X −1 (I)).. F . .1) Proof. In this case we have ∀ ω ∈ Ω. In ∈ B(E). Since any positive Borel functions is the limit of an increasing sequence of positive simple functions. the conclusion follows from the monotone convergence theorem. it is equal to 1 if ω ∈ I to 0 if ω ∈ I. l So. Consequently. . (1) 1 I (ω) is the characteristic function of I. The law (or image measure or pushforward measure) of X is the probability measure X# P on (E.1) holds for all simple functions ϕ of the form n ϕ= i=1 ci 1 I i .. ϕ(X(ω))P(dω) = P(X −1 (I)) = X# P(I) = Ω E ϕ(x)X# P(dx)... Theorem 1.4 Chapter 1 By an Evalued random variable in (Ω. (1.. Sometimes we shall use the notation X# P = PX .. ω → X(ω). l / .1 Let X be an Evalued random variable in (Ω. Let ﬁrst ϕ = 1 I with I ∈ B(E) l ϕ(X(ω)) = 1 X −1 (I) (ω). F ) we mean a mapping X : Ω → E. (1. c1 . l with n ∈ N. Let moreover ϕ : E → R be a nonnegative Borel function. ∀ I ∈ B(E). such that I ∈ B(E) ⇒ X −1 (I) ∈ F .
... 2. h µ(dx). Fi . ∀ h ∈ H. Pn . B(H)).. ∀ h ∈ H.1. We shall write m= H xµ(dx). h µ(dx). Assume that µ has ﬁnite ﬁrst momentum.Gaussian measures 5 1. Pi ). i=1 For any R = n Ai we deﬁne i=1 n P(R) := i=1 Pi (Ai ). n... 1. x2 µ(dx) < +∞. One can show that P can be uniquely extended to a probability measure on (Ω.. ∀ h ∈ H. a set of the form R = n Ai where i=1 Ai ∈ Fi . n. is continuous since F (h) ≤ H xµ(dx) h.. by deﬁnition. A meai=1 surable rectangle of Ω is. be probability spaces. . i = 1.2 1. P2 .. it is denoted by n Fi . H Then the linear functional F : H → R deﬁned as F (h) = H x.2 Product measures Let (Ωi .1 Probability measures in Hilbert spaces Mean and covariance Let µ be a probability measure on (H.. h = H x. i = 1. xµ(dx) < +∞. . The σalgebra generated by all measurable rectangles is called the product σalgebra of Fi . m is called the mean of µ. H . F ) which is called the product probability of P1 . Fn . Set Ω = n Ωi . .2.. Assume now that the second moment of µ is ﬁnite.. By the Riesz representation theorem there exists m ∈ H such that m..
k) ≤ H x − m2 µ(dx) h k. g. It is also possible to deﬁne traceclass operators which are not symmetric. Interscience. A symmetric and positive operator Q ∈ L(H) is said to be of trace class if ∞ Tr Q : = k=1 Qek . x − m k. Proof. In order to state the next result we need the concept of trace class operator. One can show that any trace class operator Q is compact and that Tr Q is the sum of its eigenvalues repeated according to their multiplicity. ∀ h. Therefore there is a unique linear bounded operator Q ∈ L(H) such that Qh. positive and of trace class. k = H h. by the monotone convergence theorem and the Parseval identity. (2) Proposition 1. ek 2 µ(dx). Then we have Qek . k ∈ H. ek < +∞ for one (and consequently for any) complete orthonormal system (ek ). x − m k.6 Chapter 1 (so that the ﬁrst one is ﬁnite as well). Schwartz. G is continuous since G(h. To prove that Q is of trace class choose a complete orthonormal system (ek ) in H. Linear Operators. 1964. ek = H  x − m.2 The covariance operator Q of µ is symmetric. N. see e. ek 2 µ(dx) = H H x − m2 µ(dx) < +∞. Part II.T. Dunford and J. we ﬁnd that ∞ Tr Q = k=1 (2)  x − m. k ∈ N. x − m µ(dx). . Therefore. Symmetry and positivity of Q are clear. k ∈ H. ∀ h. ∀ h. k ∈ H. Q is called the covariance of µ. k) = H h. x − m µ(dx). but we shall not need in what follows. Let us consider the bilinear form G : H × H → R deﬁned as G(h.
Let C ⊂ H be closed and let (ϕn ) ⊂ Cb (H) be such that (i) lim ϕn (x) = 1 C (x) for all x ∈ H. B(H).3 Let µ. ν ∈ P(H) be such that ϕ(x)µ(dx) = H H ϕ(x)ν(dx). Proof.2 Finite dimensional projections of measures We are given a probability measure µ ∈ P(H).h µ(dx). (1. We ﬁnally deﬁne the Fourier transform µ of a probability measure µ setting µ(h) = H ei x. x ∈ H.2. Let (ek ) be a complete orthonormal system in H. 1. symmetric operators in 1 H of trace class.3) We have limn→∞ Pn x = x for all x ∈ H. Proposition 1. Thus µn is a probability measure on (Pn (H). ∀ h ∈ H. ek ek . For any n ∈ N we consider the projection Pn : H → Pn (H) deﬁned as n Pn x = k=1 x. l n→∞ .4) Then µ = ν. We want now to show that µ is determined by the sequence (µn ). We shall also consider µn as a probability measure on (H. For this we ﬁrst need the following result. setting µn (I) = µn (I ∩ Pn (H)). (1. µ). (1. ∀ ϕ ∈ Cb (H). µn ). B(Pn (H)).2) One checks easily that µ : H → C is continuous. for all ϕ ∈ Cb (R). For any n ∈ N we consider the measure µn := (Pn )# µ deﬁned by ϕ(Pn x)µ(dx) = H Hn ϕ(y)µn (dy).Gaussian measures 7 We shall denote by L+ (H) the set of all positive. ∀I ∈ B(H).
A sequence (ϕn ) ⊂ Cb (H) fulﬁlling (i) and (ii) is provided by. Now. . Pn (H) Since (Pn )# µ = (Pn )# ν by assumption.4 Let µ.8 (ii) ϕn 0 Chapter 1 ≤ 1 for all ∈ N. C) if d(x. we conclude that ϕ(x)µ(dx) = H H ϕ(x)ν(dx) for all ϕ ∈ Cb (H). 1 1 − n d(x.3 we have µ = ν. Then. Therefore. Proposition 1. We can now prove the announced result. Proposition 1. by the dominate convergence theorem it follows that lim ϕn dµ = lim H n→∞ n→∞ ϕn dν = µ(C) = ν(C). using the dominated convergence theorem and the change of variables formula. we have ϕ(x)µ(dx) = lim H n→∞ ϕ(Pn x)µ(dx) = lim H n→∞ ϕ(ξ)((Pn )# µ)(dξ) Pn (H) and ϕ(x)ν(dx) = lim H n→∞ ϕ(Pn x)ν(dx) = lim H n→∞ ϕ(ξ)((Pn )# ν)(dξ). H Since closed sets generate the Borel σ–algebra of H this implies that µ = ν. ν ∈ P(H) be such that µ(h) = ν(h) for all h ∈ H. As an application of Proposition 1. C) ≥ n . C) ≤ n ϕn (x) = 1 0 if d(x. 1 if x ∈ C. Let ϕ ∈ Cb (H). in view of Proposition 1. Proof. If (Pn )# µ = (Pn )# ν for any n ∈ N we have µ = ν. ν ∈ P(H).5 Let µ.4 we prove that the Fourier transform of µ determines µ. Then µ = ν.
µ(Pn h) = H (3) 9 . / If q > 0 we set 1 Nm.Pn h (Pn )# µ(dξ) = (Pn )# µ(Pn h) and ν(Pn h) = H ei x.q (R) = √ (3) 1 2πq +∞ e− −∞ (x−m)2 2q 1 dx = √ 2π +∞ −∞ e− 2 dx = 1. The conclusion follows from Proposition 1.g.q (B) = √ 2πq e− B (x−m)2 2q dx. We assume as granted the result when H is ﬁnitedimensional the general case we have by (1. B(R)) as follows. 1.3.Pn h (Pn )# ν(dξ) = (Pn )# ν(Pn h).Pn h µ(dx) = Pn (H) ei Pn ξ. then we go to the general case. If q = 0 we set Nm. for all B ∈ B(R). M´tivier. Nm. 1.3 Gaussian probability measures We ﬁrst recall the deﬁnition of Gaussian measure on (R.Gaussian measures Proof. where δm is the Dirac measure at m. 1968.Pn h ν(dx) = Pn (H) ei Pn ξ. Therefore measures (Pn )# µ and (Pn )# ν have the same Fourier tranforms and so they coincide. q) with m ∈ R and q ≥ 0 we deﬁne a probability measure Nm. e . δm (B) = 0 if m ∈ B. deﬁned for all B ∈ B(R) by 1 if m ∈ B.1 Gaussian probability measures in R For any pair of real numbers (m.1) for any h ∈ H and n ∈ N.0 = δm .q is a probability measure since Nm. M. B(R)). In ei x.q on (R. x2 See e. Notions fondamentales de la th´orie des probabilit´es.4. Dunod e e e Universit´.
z ∈ Rn . Na. x − a µ(dx) = Qy. it is left to the reader. Rn y. en ) be an orthonormal basis on Rn such that Qek = λk ek . Then we have xµ(dx) = m. Therefore m is the mean and Q the covariance operator of Na. h ∈ Rn .q (dx) = √ (x−m)2 1 e− 2q dx. mn ) ∈ Rn and any Q ∈ L+ (Rn )...q (h) := R eihx Nm. k = 1.Q for any m = (m1 . . B(Rn )) by setting Nm. Rn y. x − a z.Q .. B(R) and Nm. . The proof of the following proposition is easy..6 Let m ∈ Rn . 1 2 (1. its Fourier transform is given by Nm. 2πq When m = 0 we shall write for short Nq instead N0.q (dx) = eimh− 2 qh . Then we deﬁne a probability measure Na.λk .10 Chapter 1 If q > 0.Q (dx) = 1 (2π)d det Q e− 2 1 Q−1 (x−a).Q is absolutely continuous with respect to the Lebesgue measure in Rn and we have Na. Q ∈ L+ (Rn ) and µ = Nm.q .x µ(dx) = ei a. Let Q ∈ L+ (Rn ) and let (e1 . When m = 0 we shall write NQ instead of Nm.Q (h) := Rn ei h..h .5) 1.Q on (Rn .Q = ×N k=1 n mk . Nm. Moreover.. n. Finally.Q for short. .. Moreover the Fourier tranform of Na.Q .x−a dx. for some λk ≥ 0.q is absolutely continuous with respect to the Lebesgue measure 1 (dx) = dx in (R. .Q is given by Na...3. h ∈ R.h − 2 1 Qh. if the determinant of Q is positive.2 Gaussian probability measures in Rn We are going to deﬁne a Gaussian measure Nm. z . It is easy to see that m is the mean and q the covariance of Nm.q . Proposition 1.
ek ek → ( x. More precisely. ek ek .6) One can show that such a measure does exist Proposition 1. An introduction to inﬁnitedimensional analysis. 2006. covariance Q and Fourier transform given by Nm.Q the probability measure on 1 (H.Gaussian measures 11 1. Da Prato. en ..Q ∈ P(H). Pn x = k=1 x. ∀x∈H and identify Pn (H) with Rn through the isomorphism.3. e1 . . B(H)) of mean m. h ∈ H.λk . (4) (1. .7 Prove that µn = (Pn )# µ = ×N i=1 n mk . see e.h .Q (h) = ei m. G. it is unique thank’s to 1.3 Gaussian probability measures in H Let m ∈ H and Q ∈ L+ (H). Exercise 1.. For any n ∈ N we set mn := m. Show that the Fourier transform of µn is given by µn (h) = ei (4) Pn k=1 1 mk hk − 2 e Pn k=1 λk h2 k .g. x.3. n Pn (H) → R .h − 2 1 Qh. given µ = Nm. en ).5. SpringerVerlag..4 Computation of some Gaussian integrals To compute some integrals with respect to a Gaussian measure µ = Nm. x = k=1 n x. n ∀ k ∈ N. Since Q is compact there exists an orthonormal complete system (ek ) in H and a sequence of nonnegative numbers (λk ) such that Qek = λk ek . Berlin. .Q in an inﬁnite dimensional Hilbert space H it is useful to reduce the computation to integrals on a sequence (Hn ) of ﬁnite dimensional vector spaces convergent to H and then to let n → ∞. We denote by Nm. we shall proceed as follows. Hint.
m .7 n e H ε P x2 2 n µ(dx) = Pn (H) e ε P ξ2 2 n µn (dξ) = k=1 R e 2 ξk Nmk . λ1 (1. . the linear operator 1 − εQ is invertible and (1 − εQ)−1 is bounded. For any n ∈ N we have. Proposition 1.9 Let ε ∈ R.λk (dξk ). ek ek . k=1 (1 − ελk ) > 0. as easily checked. if ε < 1 . ε 2 Since Pn x2 ↑ x2 as n → ∞ and. Then we have ε −1 [det(1 − εQ)]−1/2 e 2 (1−εQ) m. (1 − εQ) x = 1 − ελk k=1 In this case we can deﬁne the determinant of (1 − εQ) by setting n ∞ det(1 − εQ) : = lim Exercise 1. k=1 Hint. x ∈ H.λk (dxk ) = √ R ε 2 1 −ε e 2 1 − ελk m2 k 1−ελk .8 Prove that ∞ n→∞ (1 − ελk ) := k=1 (1 − ελk ). by an elementary computation. We have in fact. 1 To formulate the next result notice that for any ε < λ1 .12 Chapter 1 We shall assume (which is always true after a rearrangement) that λ1 ≥ λ2 ≥ · · · λn ≥ · · · . ε 2 e 2 x µ(dx) = H +∞. taking into account Exercise 1. otherwise. Write log ∞ ∞ (1 − ελk ) k=1 = k=1 log(1 − ελk ) ∞ k=1 and show that the series is convergent since λk < +∞. e 2 xk Nmk . ∞ 1 −1 x. the conclusion follows from the monotone convergence theorem.7) Proof.
Gaussian measures Exercise 1.10 Prove that for all m ∈ N Jm :=
H
13
x2m µ(dx) < ∞
and compute Jm . Hint. Notice that Jm = 2m F (m) (0), where F (ε) =
H
e 2 x µ(dx),
ε
2
ε > 0.
Proposition 1.11 We have e h,x µ(dx) = e a,h e 2
H
1
Qh,h
, h ∈ H.
(1.8)
Proof. For any ε > 0 we have e h,x ≤ ex h ≤ eεx e ε h . Choosing ε <
1 , λ1
2 1 2
we have, by the dominated convergence theorem, that
n→∞
e h,x µ(dx) =
H
lim
e h,Pn x µ(dx) = lim
H
1
n→∞
e h,Pn ξ µn (dx)
Pn (H)
1
=
n→∞
lim e Pn m,h e 2
Pn Qh,h
= e m,h e 2
Qh,h
.
1.3.5
The Cameron–Martin space
We are given a Gaussian measure µ = NQ , where Q ∈ L+ (H). We say that 1 µ is non degenerate if Ker Q := {x ∈ H : Qx = 0} = {0}. Thus, if H is ﬁnitedimensional µ is non degenerate if and only if det Q > 0. Assume now that H is inﬁnitedimensional and that µ is non degenerate. We denote by (ek ) a complete orthonormal system in H such that Qek = λk ek , k ∈ N, where (λk ) are the eigenvalues of Q and we set xk = x, ek , k ∈ N. We notice that the inverse Q−1 of Q (which is well deﬁned since Ker Q = {0}) is not continuous because, Q−1 ek = 1 ek , λk k∈N
and λk → 0 as k → ∞. Consequently, recalling the closed graph theorem, we see that the range Q(H) does not coincide with H. However, it is dense in H as the following lemma shows.
14 Lemma 1.12 Q(H) is a dense subspace of H.
Chapter 1
Proof. In fact if x0 is an element of H orthogonal to Q(H), we have Qx, x0 = x, Qx0 = 0, ∀ x ∈ H, which yields Qx0 = 0, and so x0 = 0 because Ker(Q) = {0}. It is useful to introduce the operator Q1/2 deﬁned as
∞
Q1/2 x =
k=1
λk x, ek ek ,
x ∈ H.
Its range Q1/2 (H) is called the Cameron–Martin space of the measure µ. Arguing as before we see that Q1/2 (H) is a subspace of H diﬀerent of H and dense in H. Moreover it is clear that x ∈ Q1/2 (H) if and only if,
∞
λ−1 x2 < +∞. k k
k=1
It is important to notice that the measure of the Cameron–Martin space is zero. Proposition 1.13 We have µ(Q1/2 (H)) = 0. Proof. For any n, k ∈ N set
∞
Un = and
y∈H:
h=1
2 λ−1 yh < n2 h
= {y ∈ Q1/2 (H) : Q−1/2 y < n},
2k
Un,k =
1/2
y∈H:
h=1
2 λ−1 yh < n2 h
.
Clearly Un ↑ Q (H) as n → ∞, and for any n ∈ N, Un,k ↓ Un as k → ∞. So, it is enough to show that µ(Un ) = lim µ(Un,k ) = 0.
k→∞
(1.9)
We have in fact µ(Un,k ) =
{y∈H:
P2k
−1 2 2 h=1 h=1 λh yh <n }
×N
2k
λk (dyk ),
Gaussian measures which, setting zh = λh
−1/2
15 yh is equivalent to NI2k (dz),
µ(Un,k ) =
{z∈R2k :z<n}
where I2k is the identity in R2k . Let us compute µ(Un,k ). We have µ(Un,k ) µ(Un,k ) = = µ(H) Therefore 1 µ(Un,k ) = (k − 1)! and (1.9) follows.
n2 /2 n − r2 2k−1 e 2r dr 0 2 +∞ − r 2k−1 e 2r dr 0
=
n2 /2 −ρ k−1 e ρ dρ 0 . +∞ −ρ k−1 e ρ dρ 0
e ρ
0
−ρ k−1
1 dρ ≤ (k − 1)!
n2 /2
ρ
0
k−1
1 dρ = k!
n2 2
k
,
16 Chapter 1 .
∀ h ∈ H. F . P) be a probability space. k P(dω).1 We say that X# P is a Gaussian random variable if X# P is a Gaussian measure. and Q(X)h.h .1 Notations Let (Ω. h X(ω) − m(X). ∀ h. h = Ω X(ω).Chapter 2 Gaussian random variables 2. k = Ω X(ω) − m(X). ∀h∈H and that m(X). h P(dω). X : Ω → H a random variable such that X(ω)2 P(dω) < ∞. that is if X# P(h) = ei m(X).h e− 2 1 Q(X)h. In this case we call m(X) the mean and Q(X) the covariance of X. Ω We denote by X# P the law of X. By the change of variables formula it follows that the Fourier transform of X# P is given by X# P(h) = Ω ei X(ω). 17 .h P(dω). Deﬁnition 2. by m(X) the mean of X# P and by Q(X) the covariance of X# P. k ∈ H. ∀ h ∈ H. H a separable Hilbert space.
. .X(ω) P(dω) = eiam(Xk ) e− 2 a 1 2 Q(X ) k = ei aek .. 2. P). Xn be real random variables in (Ω. i.. Xn are real Gaussian random variables. n we have Q(X)j. ω ∈ Ω. Then X = (X1 .. Xn are real Gaussian random variables... n and a ∈ R we have eiaXk (ω) P(dω) = Ω Ω 1 ei aek .. .. j = 1. m(X) is a vector of Rn denoted by (m(X)1 .. . n. F . k = 1. m(X)n ) and Q(X) is a n × n matrix denoted Q(X)i.. More precisely. ek = Ω (Xj (ω) − mj (Xj ))(Xk (ω) − mk (Xk ))P(dω)..... F ..18 Chapter 2 Example 2... . k = 1.m(X) e− 2 a2 Q(X)ek . . Xn (ω)). In particular..2. .. let (e1 . n.. if j = k we ﬁnd Q(X)k... .. Example 2. Xn ) is a Rn valued random variable.... Xn be real random variables on (Ω. Consider the Rn valued random variable X(ω) = (X1 (ω). Then X1 . . So.. . .. Xn ) is not necessarily Gaussian. then X = (X1 .. . In fact if k = 1. . . X1 . . ek = Ω Xk (ω)P(dω) = m(Xk ) and for any j.. .2 Let n ∈ N.4 Let n ∈ N and let X1 ....ek .. if conversely X1 .1 Independent real variables Deﬁnition 2. en ) be the canonical basis in Rn .3 Assume that X = (X1 .. Then for any k = 1.. n we have m(X)k = m(X).j . Xn ) is a ndimensional Gaussian random variable. P).. 2.. ..k = Q(X)ej .k = Q(Xk ).2 Independence In this section we introduce the basic concept of independence.... Notice that....
... ξn ) ∈ Rn ...... l i = 1.... be real independent random variables in (Ω. Proof. Assume conversely that (2. Conversely. . . But this follows immediately setting in (2. Xn are independent.. To prove independence of X1 .... .. They are called independent if Xi1 ..random variables We say that X1 . if (2.. ϕn positive Borel. n ∈ N. . ξn ) = ϕ1 (ξ1 ) · · · ϕk (ξn ).. Xin are independent for any choice of n and of positive integers i1 < i2 < · · · < in .. A necessary and suﬃcient condition for the independence is provided by the following proposition.. . Xn . In ∈ B(R)...1) holds for any choice of functions ϕ1 . ... . Let (Xi ) be a sequence of real random variables.. . Set X = (X1 .. Proposition 2. . . Then we have ϕ1 (X1 (ω)) · · · ϕn (Xn (ω))P(dω) Ω (2. ... Xn ) and let ψ : Rn → R be deﬁned as ψ(ξ1 .. (ξ1 .1) = Ω ϕ1 (X1 (ω))P(dω) · · · Ω ϕn (Xn (ω))P(dω). ϕn be Borel positive functions. ... . Let moreover ϕ1 . ∀ I1 . Xn it is enough to show that (X# P)(I1 × · · · × In ) = ((X1 )# P)(I1 ) · · · ((Xn )# P)(In ).. n... . Xn . . ϕn ..1) holds for any choice of positive Borel functions ϕ1 . taking into account the independence of X1 . . P). F .5 Let X1 ... then X1 . ϕ1 (X1 (ω)) · · · ϕn (Xn (ω))P(dω) = Ω Ω ψ(X(ω))P(dω) = Rn ψ(ξ)(X# P)(dξ) = R ϕ1 (ξ1 )((X1 )# P)(dξ1 ) · · · R ϕk (ξn )((Xn )# P)(dξn ) = Ω ϕ1 (X1 (ω))P(dω) · · · Ω ϕn (Xn (ω))P(dω).. Then by the change of variable formula we have.. . Xn are independent if X# P = 19 ×(X ) j=1 n j # P.1) ϕi = 1 Ii .
P) and let X = (X1 . . ... Proposition 2. F . Xn be real independent random variables in (Ω.. . ... . Xn be real independent random variables in (Ω.9 Show that sets A1 ..7 Let X1 .. jk less or equal to n... P) be a probability space and A1 .. ∀ h = (h1 .8 Let (Ω. n and k diﬀerent positive integer j1 . An ∈ F .. An are independent if and only if P(Aj1 ∩ · · · ∩ Ajk ) = P(Aj1 ) × · · · × P(Ajk ).. ... Show that X1 · · · Xn dP = Ω Ω X1 dP × · · · × Ω Xn dP and V (X1 + · · · + Xn ) = V (X1 ) + · · · + V (Xn ). Then X1 .. Then the covariance matrix Q(X) is diagonal.6 Let X1 ... . Xn ).. Proof.20 Chapter 2 Exercise 2... .. Proposition 2.... We have in fact (by Exercise 2. We say that the sets A1 . . P)...j = Ω (Xi (ω) − mi (X))(Xj (ω) − mj (X))P(dω) (Xj (ω) − mj (X))P(dω) = 0.. . Xn )... hn ) ∈ Rn . Ω = Ω (Xi (ω) − mi (X))P(dω) The converse of Proposition 2. F . . l l Exercise 2..... for all k = 1. Xn be real random variables in (Ω. . n Q(X)i. F . j = 1. An are independent if the random variables 1 A1 .. Xn are independent if and only if n X# P(h) = k=1 (Xk )# P(hk ).. ..10 Let X1 . P) and let X = (X1 ... Deﬁnition 2.. F .6) for i.. ..10 does not hold in general.. 1 An are so. The following useful result is left to the reader as an exercise. .
1 . Proof. . In fact..Q with m ∈ H and Q ∈ L+ (H).h e− 2 =e i m(X).. Then... . Xn are independent the conclusion follows from Proposition 2. ...random variables 21 2. . Xn are real random variables and that X = (X1 .h e− 2 n 1 Pn k=1 Q(X)k...k h2 k e 1 −2 Pn k=1 Q(Xk )h2 k = i=1 (Xk )# P(h).... .. for each h = (h1 . Then X = (X1 . Xn be real random variables in (Ω. If X1 . ... P) and let X = (X1 . Assume now that Q(X) is diagonal.. . . .11. taking into account the independence of (X1 .7 it is enough to show that n X# P(h) = i=1 (Xk )# P(h). . 2.2.. . Xn are independent if and only if Q(X) is diagonal. 1 2 Proposition 2. hn ) ∈ H.. Then X1 ..11 Assume that X1 .. hn ) ∈ Rn . We have in fact X# P(h) = ei m(X). Xn are independent Gaussian random variables....h 1 Q(X)h... By Proposition 2. Xn ). F .. B(H).12 Assume that X1 .. Xn ) is Gaussian. where H is a separable Hilbert space and µ = Nm.h = ei m(X).3 Gaussian random variables deﬁned in a Hilbert space We now consider the case when (Ω.... P) coincides with (H.. Proposition 2. µ).. F ..2 Independent Gaussian random variables Let X1 . Proof.. Xn ). n X# P(h) = Ω e i(X1 (ω)h1 +···+X1 (ω)hn ) P(dω) = k=1 Ω 2 eiXk (ω)hk P(dω) = ei(m(X1 )h1 +···+m(Xn )hn ) e− 2 (Q(X1 )h1 +···+Q(Xn )hn ) . Xn ) is Gaussian. let h = (h1 .
22
Chapter 2
2.3.1
Aﬃne changes of variables
Let b ∈ K and A ∈ L(H, K) where K is another separable Hilbert space. Let us consider the aﬃne transformation T (x) = Ax + b, x ∈ H.
Proposition 2.13 T is a Gaussian random variable and its law T# µ is given by NAa+b,AQA∗ , where A∗ is the transpose of A. Proof. We have in fact ei k,y T# µ(dy) =
K H
ei k,T (x) µ(dx) =
H
1
ei k,Ax+b µ(dx)
AQA∗ k,k
= ei k,b
H
ei A
∗ k,x
µ(dx) = ei k,Aa+b e− 2
, k ∈ K.
Example 2.14 Let µ = Nm,Q and n ∈ N, f1 , ..., fn ∈ H. Let F : H → Rn be deﬁned as F (x) := ( x, f1 , ..., x, fn ), x ∈ H. Then by Proposition 2.13 F is a Gaussian random variable with mean m(F ) and covariance Q(F ) given by, m(F ) = F (m) = ( m, f1 , ..., m, fn ) and Q(F ) = F QF ∗ . On the other hand, the linear operator F ∗ : Rn → H is given by
n
F (ξ) =
k=1
∗
fk ξk ,
∀ ξ = (ξ1 , ..., ξn ) ∈ Rn .
Therefore QF (ξ) =
∗
n
Qfk ξk ,
k=1
∀ ξ = (ξ1 , ..., ξn ) ∈ Rn
n
and F QF ∗ (ξ) =
n
Qfk ξk , f1
k=1
, ...,
k=1
Qfk ξk , fn
random variables so that Q(F )h,k = Qfh , fk . Therefore, F1 , ..., Fn are independent if and only if Qfh , fk = 0, if h = k. h, k = 1, ..., n,
23
(2.2)
2.4
The white noise function
In order to deﬁne the white noise function (which will play an important role in what follows), we shall deal with equivalence class of random variables (rather than random variables), which we brieﬂy discuss in the next subsection.
2.4.1
Equivalence classes of random variables
Let (Ω, F , P) be a probability space and let H be a separable Hilbert space. We denote by R(H) the set of all Hvalued random variables. Deﬁnition 2.15 We say that X, Y ∈ R(H) are equivalent (and write X ∼ Y ) if P({ω ∈ Ω : X(ω) = Y (ω)}) = 1. One can easily check that X ∼ Y, X, Y ∈ R(H) is an equivalence relation, so that the set R(H) is disjoint union of equivalences classes. We notice that if X ∼ Y then the laws of X and Y coincide. In fact set K = {ω ∈ Ω : X(ω) = Y (ω)}, so that P(K) = 0. Since for any I ∈ B(H) we have X −1 (I) ⊂ Y −1 (I) ∪ K, it follows that P(X −1 (I)) ≤ P(Y −1 (I)) and, exchanging X and Y we see that P(X −1 (I)) = P(Y −1 (I)). Consequently, all random variables belonging to a ﬁxed equivalence class ˜ ˜ X have the same law, which is called the law of X. In the following we shall not distinguish between a random variable X ˜ and the equivalence class X including X, except when needed.
24
Chapter 2
By Lp (Ω, F , P; H), p ≥ 1, we mean the space of all equivalence class of random variables X : Ω → H such that X(ω)p P(dω) < +∞.
Ω
Lp (Ω, F , P; H), endowed with the norm
1/p
X
Lp (Ω,F ,P;H)
=
Ω
X(ω)p P(dω)
,
is a Banach space. We shall write Lp (Ω, F , P; H) = Lp (Ω, P; H) for brevity. We prove now that the limit of a convergent sequence in L2 (Ω, P; H) of Gaussian random variables is Gaussian. Proposition 2.16 Let (Xn ) ⊂ L2 (Ω, P; H) be a sequence of Gaussian random variables convergent to X in L2 (Ω, P; H). Then X is a Gaussian random variable and m(X), h = lim m(Xn ), h , h ∈ H,
n→∞
and Q(X)h, k = lim Q(Xn )h, k ,
n→∞
h, k ∈ H.
Proof. Since Xn → X in L2 (Ω, P; H) we have
n→∞
lim m(Xn ), h = lim
n→∞
Xn (ω), h P(dω) =
Ω Ω
X(ω), h P(dω) = m(X), h
and
n→∞
lim Q(Xn )h, k
=
n→∞
lim
Xn (ω) − m(Xn ), h
Ω
Xn (ω) − m(Xn ), k P(dω)
=
Ω
X(ω) − m(X), h
X(ω) − m(X), k P(dω) = Q(X)h, k .
Let us show now that X is a Gaussian random variable. We have in fact ei x,h (X# µ)P(dy) =
H Ω
1
ei X(ω),h P(dω) = lim = ei m(X),k e− 2
1
n→∞
ei Xn (ω),h P(dω)
Ω
= lim ei m(Xn ),h e− 2
n→∞
Q(Xn )h,h
Q(X)h,h
.
Let us deﬁne a mapping W : Q1/2 (H) → C(H). x ∈ Q1/2 (H). µ) which we denote still by W and call the white noise function.e. Since Q1/2 (H) is dense in H. Remark 2. Q−1/2 z1 x.4. z2 . where Wz (x) = x.13 .17 For all z1 .18 Given z ∈ H (not belonging to Q1/2 (H)) it would be tempting to deﬁne the random variable Wz by setting. x. ∀ x ∈ H. the mapping W can be uniquely extended as a mapping from H into L2 (H. z → Wz (2. Lemma 2. Q−1/2 z2 µ(dx) = QQ−1/2 z1 . H k ∈ N. y µ a.3) Proof. Wf is linear in the sense that for all α. Wz (x) = Q−1/2 x.2 Deﬁnition of the white noise function In this section we assume that the Hilbert space H is inﬁnite dimensional and consider a non degenerate Gaussian measure µ = NQ in H (Ker (Q) = {0}). QQ−1/2 z2 = z1 .. We have in fact Wz1 (x)Wz2 (x)µ(dx) = H H x. by Proposition 1. Since Q is compact there exists a complete orthonormal basis (ek ) on H and a sequence of positive numbers (λk ) such that Qek = λk ek . Q−1/2 z . However this deﬁnition is meaningless because µ(Q1/2 (H)) = 0. z2 ∈ Q1/2 (H) we have Wz1 (x)Wz2 (x)µ(dx) = z1 . z . z2 .random variables 25 2. β ∈ R we have Wf (αx + βy) = αWf (x) + βWf (y). Here Q1/2 (H) is the Cameron–Martin space and C(H) the space of all real continuous functions on H.
.. zn in H.. zn are mutually orthogonal... .19 is important. ∀ η ∈ R.. We have to show that eiηWz (x) µ(dx) = e− 2 η H 1 2 z2 . So.... k = 1.. n.20 Let n ∈ N. Then (Wz1 . z1 . Wzn ) is an ndimensional Gaussian random variable with mean 0 and covariance operator Qz given by (Qz )h.k=1 zj . Then Wz is a real Gaussian random variable with mean 0 and covariance z2 . . Wzn are independent if and only if z1 .. Proposition 2. The following generalization of Proposition 2.4) The random variables Wz1 .. Then we have by the dominated convergence theorem.x ) µ(dx) = lim j→∞ ei x.Q H 1 j −1/2 (ξ j j 1 z1 +···+ξn zn ) µ(dx) 2 1 = lim e− 2 ξ1 z1 +···+ξn zn  = e− 2 ξ1 z1 +···+ξn zn  = e− 2 j→∞ j 2 1 Pn j. (2. (zj ) be n sequences in Q1/2 (H) convergent respectively to z1 ..x +···+ξn Q−1/2 zn . .. n 1 Proof...19 Let z ∈ H. .26 Chapter 2 Proposition 2. . . by the dominated convergence theorem. Proof. .zk ξj ξk . Then. the conclusion follows.x µ(dx) = lim e− 2 η n→∞ 1 2 z n 2 = e− 2 η 1 2 z2 ....k = zh . . Let (zj ). zk . zn ∈ H.. h.. that ei(ξ1 Wz1 (x)+···+ξn Wzn (x)) µ(dx) = lim H j→∞ ei(ξ1 H j j Q−1/2 z1 . Let (zn ) ⊂ Q1/2 (H) be a sequence such that zn → z in H. we have eiηWz (x) µ(dx) = lim H n→∞ eiη Q H −1/2 z n .
Chapter 3 Brownian Motion
3.1 Stochastic Processes
We are given a probability space (Ω, F , P). We denote by P∗ the outer measure of P. We recall that a null set of Ω is a set of outer measure zero. For any integrable real random variable F we note E(F ) =
Ω
F (ω)P(dω).
So, in particular we have F# P(I) = E(1 I (F )), l ∀ I ∈ B(R).
We say that a property π concerning elements of Ω holds Pa.s. if the set where π does not hold is a null set. Deﬁnition 3.1 A family X = (X(t))t≥0 of real random variables in (Ω, F , P) is called a real stochastic process in [0, +∞). For any ω ∈ Ω, X(·, ω) is called a trajectory of X. • X is Gaussian if for any n ∈ N and any 0 ≤ t1 < · · · < tn the ndimensional random variable (X(t1 ), ..., X(tn )) is Gaussian. • X is continuous if X(·, ω) is continuous Pa.s. • X is pmean continuous, p ≥ 1, if (i) X(t) is pintegrable for any t ≥ 0. (ii) We have
t→t0
lim E[X(t) − X(t0 )p ] = 0, 27
∀ t0 ≥ 0.
(3.1)
28
Chapter 3
We notice that a pmean continuous process is not continuous in general. We say that two stochastic processes X and Y are equivalent if for all t ≥ 0 we have X(t, ω) = Y (t, ω), Pa.s..
When X and Y are equivalent we also say that Y is a version of X (or that X is a version of Y ).
3.2
Brownian motion
Deﬁnition 3.2 A real Brownian motion B = (B(t))t≥0 on (Ω, F , P) is a real stochastic process such that (i) B(0) = 0 and if 0 ≤ s < t, B(t) − B(s) is a real Gaussian random variable with law Nt−s . (ii) If 0 < t1 < ... < tn , the random variables, B(t1 ), B(t2 ) − B(t1 ), · · · , B(tn ) − B(tn−1 ) are independent. We express condition (ii) by saying that B is a process with independent increments. Lemma 3.3 Let t, s > 0. Then E[B(t)(B(s)] = min{t, s}. Proof. Let for instance t > s. Then we have E[B(t)B(s)] = E[(B(t) − B(s))B(s)] + E[B 2 (s)]. On the other hand, B(t) − B(s) is independent of B(s) so that E[(B(t) − B(s))B(s)] = E[B(t) − B(s)]E[B(s)] = 0. Since the law of B(s) is Ns we conclude that E[B(t)B(s)] = s as required. (3.2)
Brownian motion
29
3.2.1
Construction of a Brownian motion
Consider the probability space (H, B(H), µ), where H = L2 (0, +∞) and µ = NQ , Q being an arbitrary (but ﬁxed) non degenerate Gaussian measure in H. Deﬁne B(t) = W1l[0,t] , t ≥ 0, (3.3) where 1 [0,t] (s) = l 1 if s ∈ [0, t], 0 otherwise,
and W is the white noise function deﬁned in Chapter 2. More precisely, for any t ≥ 0 we choose an arbitrary element in the equivalence class of B(t) which we still denote by B(t). Clearly, for any t ≥ 0, B(t) is a Gaussian random variable Nt and for any t > s ≥ 0, B(t) − B(s) = W1l(s,t] is a Gaussian random variable Nt−s . So, B fulﬁlls Deﬁnition 3.2(i). Let us prove (ii). Since the system of elements of H, (1 [0,t1 ] , 1 (t1 ,t2 ] , ..., 1 (tn−1 ,tn ] ), l l l is orthogonal, we have by Proposition 2.20 that the random variables B(t1 ), B(t2 ) − B(t1 ), · · · , B(tn ) − B(tn−1 ) are independent. Thus (ii) is proved as well.
3.2.2
Some properties of a Brownian motion
Proposition 3.4 Let B(t), t ≥ 0, be a Brownian motion on (Ω, F , P). Then B is a Gaussian process. Moreover, if 0 < t1 < ... < tn the law of (B(t1 ), ..., B(tn )) is given by P((B(t1 ), ..., B(tn )) ∈ I) = (2π)−n/2 (t1 (t2 − t1 ) × · · · × (tn − tn−1 ))−1/2
I
e
− 2t1 −
1
η2
(ηn −ηn−1 )2 (η2 −η1 )2 −·− 2(t −t 2(t2 −t1 ) n n−1 )
dη,
(3.4) for all I ∈ B(R ).
n
Proof. Let 0 < t1 < ... < tn and set X := (B(t1 ), B(t2 ) − B(t1 ), ..., B(tn ) − B(tn−1 )) Z := (B(t1 ), ..., B(tn )).
F .. Now.11 it follows that X is a ndimensional Gaussian random variable with mean 0 and covariance operator Q(X) = diag (t1 .4). the conclusion follows. . Then B is pmean square continuous for all p ≥ 1. F . .. Moreover. ηn − ηn−1 ). m!2m Therefore lim E(B(t) − B(t0 )2m ) = 0 t→0 and the conclusion follows... m ∈ N. x1 + · · · + xn ). we have (Q(Z)) η. T −1 and so. we have E(B(t) − B(t0 )2m ) = R ξ2m Nt−t0 (dξ) = (2m)! (t − t0 )m . since T −1 η = (η1 . we have det Q(Z) = det Q(X) = t1 (t2 − t1 ) × · · · × (tn − tn−1 ). consider the linear mapping T ∈ L(Rn ) deﬁned by..5 Let B(t). xn ) = (x1 . Prove that the following are Brownian motions. If I ∈ B(Rn ) we have P(Z ∈ I) = (2π)−n/2 (det Q(Z))−1/2 I e− 2 1 (Q(Z))−1 η. Therefore by Proposition 2. P).. xn ) ∈ Rn .. .. Since det T = det T ∗ = 1. B(tn ) − B(tn−1 ) are independent. t ≥ 0. It remain to show (3. tn − tn−1 ). Since B(t) − B(t0 ) is a Gaussian random variable Nt−t0 . η2 − η1 . . T (x1 .13 Z is Gaussian with mean 0 and covariance Q(Z) = T Q(X)T ∗ where T ∗ is the transpose of T . Exercise 3. Proof. t2 − t1 ... x1 + x2 . It is clear that Z = T (X). P)... ..30 Chapter 3 Since random variables B(t1 ). by Proposition 2. η = Q T −1 −1 −1 2 (ηn − ηn−1 )2 η1 (η2 − η1 )2 − ··· − − η = t1 (t2 − t1 ) (tn − tn−1 ) η. Proposition 3. as easily checked... be a Brownian motion on (Ω.6 Let B(t) be a Brownian motion in a probability space (Ω. . ... Let t > t0 ≥ 0. B(t2 ) − B(t1 ).η dη.. It is enough to show the result for p = 2m. ∀ (x1 .
Lemma 3. F . T ) with T > 0. where h > 0 is given. (iii) B3 (t) = tB(1/t). t > 0.5) is obvious. 0 We start with step functions. We have n E(Iσ (f ) ) = E j=1 n 2 f (tj−1 )2 [B(tj ) − B(tj−1 )]2 (3. 31 3. f0 . B3 (0) = 0.. where α > 0 is given.7 We have T E 0 f (s)dB(s) =0 (3. (ii) B2 (t) = αB(α−2 t). be a Brownian motion in (Ω.Brownian motion (i) B1 (t) = B(t + h) − B(h). Let 0 = t0 < t1 < · · · < tn = T . l Then deﬁne T n f (s)dB(s) := 0 j=1 ftj−1 (B(tj ) − B(tj−1 )). We want to deﬁne the stochastic integral: T f (s)dB(s). . fn−1 ∈ R and set n f= j=1 tj−1 1 (tj −tj−1 ] . t ≥ 0.5) and T 2 n t E 0 f (s)dB(s) = j=1 f (tj−1 )2 (tj − tj−1 ) = 0 f 2 (s)ds. P) and let f ∈ L2 (0. (3.3 Wiener integral Let B(t). f1 . . (iv) B4 (t) = −B(t).6).. t ≥ 0.. t ≥ 0.7) +2E j<k f (tj−1 )f (tk−1 )[B(tj ) − B(tj−1 )][B(tk ) − B(tk−1 )] . Let us prove two basic identities. Let us prove (3.6) Proof. t ≥ 0. Identity (3.
T ).6) it follows that the linear mapping I T S(0. T ). Since S(0.8) and T t E 0 f (s)dB(s) = 0 f 2 (s)ds. the equivalence class of random T variables) 0 f (s)dB(s). (3. which belongs to L2 (Ω. . T ) the linear space of all step functions. b. F . F .9 Let f ∈ L2 (0. Show that T T T E 0 f (s)dB(s) 0 g(s)dB(s) = 0 f (s)g(s)ds. c ≥ 0 we have b c c f (s)dB(s) + a b f (s)dB(s) = a f (s)dB(s). T ) → L2 (Ω. We still denote by I(f ) = 0 f (s)dB(s) this estension. T ) ⊂ L2 (0. P). is called the Wiener integral of f in [0. It is clear that for any f ∈ L2 (0. It is easy to see that if a. b ≥ 0. b We deﬁne in an obvious way the Wiener integral a f (s)dB(s) for any a. T ). is continuous. g ∈ L2 (0. T ]. T 0 f (s)dB(s) is a real Proof. Proposition 3. It is enough to prove the result for f of the form n f= i=1 fti−1 (ti − ti−1 ). P).8 Let f. T ) is dense in L2 (0. T ) it can be uniquely extended T to the whole L2 (0. T ) we have T E 0 f (s)dB(s) 2 = 0. f → I(f ) = 0 f (s)dB(s). Exercise 3.9) The random variable (more precisely. By (3. Denote by S(0. Then I(f ) = T Gaussian random variable Nq with q = 0 f (s)2 ds.32 Chapter 3 Now the conclusion follows taking into account that B(tj ) − B(tj−1 ) is a real Gaussian random variable Ntj−1 −tj and that B(tj ) − B(tj−1 ) is independent of B(tk ) − B(tk−1 ) for k = j. (3.
. Then we have ∞ Wf = 0 f (s)dB(s). . ∞).tk ] = Wf . In this case we have in fact ∞ n f (s)dB(s) = 0 k=1 ftk−1 W1l(tk−1 . ∀ t ≥ 0. B(tn ) − B(tn−1 ). · · · . Since random variables B(t1 ).tk ] . we have that I(f ) is a real Gaussian random variable Nq with n q= i=1 f 2 (ti−1 )(ti − ti−1 ). ∞) → R such that it is integrable in all interval [0.2.Brownian motion where n ∈ N. Proposition 3. k=1 Let f : [0. are independent.10) when n f= k=1 l ftk−1 1 (tk−1 . Let f ∈ L2 (0.10 We use here notations of Section 3. 0 = t0 < t1 < . B(t2 ) − B(t1 ). so that n 33 I(f ) = i=1 fti−1 (B(ti ) − B(ti−1 )). < tn−1 = T . T 0. where 0 ≤ t0 < · · · < tn . Let us introduce a stochastic process setting t F (t) = 0 f (s)ds. Example 3.1. We now show a relation between the white noise function and the Wiener integral. (3..tk ] = WPn ftk−1 1l(tk−1 .11 The process F (t). T ].10) It is enough to show (3. t ≥ 0 is pmean continuous for any p ≥ 1.
. Proposition 3. Let σ = {t0 . Pa.. n. T ]) we have T T f (s)dB(s) = f (T )B(T ) − 0 0 f (s)B(s)ds. Then by Proposition 3. k = 1..12 If f ∈ C 1 ([0. T ]) then it is possible to express the T Wiener integral 0 f (s)dB(s) in terms of a Riemann integral as the following integration by parts formula shows. We note ﬁnally. ω ∈ Ω. It follows that T σ→0 lim Iσ (f ) = f (T )B(T ) − 0 f (s)dB(s)ds.34 Chapter 3 Proof. · · · . t→t0 f 2 (s)ds. tn } ∈ Σ.9 we have that t F (t) − F (t0 ) = t0 f (s)dB(s) t t0 is a real Gaussian random variable with mean 0 and covariance Therefore q t (2m)! 2m 2 EF (t) − F (t0 ) = f (s)ds . m ∈ N and t > t0 ≥ 0.e. t1 . that if f ∈ C 1 ([0. Let p = 2m. Pa. Then we have n Iσ (f ) = k=1 n f (tk−1 )(B(tk ) − B(tk−1 )) = k=1 n (f (tk )B(tk ) − f (tk−1 )B(tk−1 )) − k=1 (f (tk ) − f (tk−1 ))B(tk ) n = f (T )B(T ) − k=1 n (f (tk ) − f (tk−1 ))B(tk ) = f (T )B(T ) − k=1 f (αk )B(tk )(tk − tk−1 )..11) Proof.. .s. where αk are suitable numbers in the interval [tk−1 . tk ]. m!2m t0 so that lim EF (t) − F (t0 )2m = 0. (3.
Proposition 3. 1/2) we have B(t) = where Yα (σ) = 0 sin πα π σ t (t − σ)α−1 Yα (σ)dσ. (3. (1) This requires a proof which is left to the reader. 1/2). Exchanging integrals . Proof. F . 0 (3. obviously.14) where α ∈ (0.14) it is enough to set σ = r(t − s) + s so that (3. t (t − σ)α−1 (σ − s)−α dσ = s π . 1). To this purpose we shall use a representation formula for B proved in the next proposition. be a Brownian motion on a probability space (Ω. be a Brownian motion on a probability space (Ω. t ≥ 0. sin πα 0 ≤ s ≤ σ ≤ t. 1 − α) = 0 π . We can now prove the result. To check (3. Then B possesses a continuous version. P). We are going to show that B possesses a continuous version. 1). F . P).14) becomes 1 (1 − r)α−1 r−α dr = β(α. (3. We start from the following elementary identity which is valid for any α ∈ (0.12) (σ − s)−α dB(s). sin πα Now since.Brownian motion 35 3. yields t σ sin πα B(t) = π dξ(t − σ) 0 α−1 0 (σ − s)−α dB(s) . B(t) = B(t) = sin πα π (1) t 0 s t s 0 dB(s) we can write (t − σ)α−1 (σ − s)−α dσ dB(s). .4 Continuity of Brownian motion Let B(t).14 Let B(t).13) Notice that the Wiener integral Yα is meaningful since α ∈ (0.13 For any α ∈ (0. Theorem 3. t ≥ 0.
15) 2m (Notice that (α − 1) 2m−1 > −1. and F is continuous as required. t ≥ 0. ω) of the stochastic process Yα which is 2mintegrable with 2m > 1/α.15 Let α ∈ (0. Now set t sin πα B(t.T . T ]. 1]. where C0 = {η ∈ C([0. t ∈ [0. By H¨lder’s inequality we have o t 2m−1 2m F (t) ≤ 0 (t − σ) 2m (α−1) 2m−1 dσ f L2m (0. t0 Let us set for ε < 2 . T ). P) such that B(·. in a probability space (Ω. Choose a version Yα (·.H) . ω) is continuous for all ω ∈ Ω. t−ε Fε (t) = 0 (t − σ)α−1 f (σ)dσ.T .11. 0 Fε is obviously continuous on [ t2 . T ]. T ] for any t0 ∈ (0. T ]. T ]. Moreover. ω)dσ. uniformly on [ t2 . F . Proof. 3. ω) is a continuous version of B thanks to the following analytic lemma. 1/2). π 0 Then B(·. H). m ∈ N with 2m > 1/α and f ∈ L2m (0. t ∈ [0. +∞)) : η(0) = 0}. Exercise 3. T ].16 Prove that B possesses an H¨lder continuous version with o any exponent β < 1/2. T . Then F ∈ C([0. we ﬁnd F (t) − Fε (t) ≤ M 2m − 1 2mα − 1 2m−1 2m εα− 2m f L2m (0. ω → B(·. ω) = (t − σ)α−1 Yα (σ. Set t F (t) = 0 (t − σ)α−1 f (σ)dσ.5 The standard Brownian motion Let us consider a Brownian motion B(t). We denote by B the mapping B : Ω → C0 .) Therefore F ∈ L∞ (0. (3. Lemma 3. H) and F is con0 tinuous at 0. This is possible in view of Proposition 3. Let us prove that F is continuous on [ t2 . ∀ t ≥ 0. .36 Chapter 3 Proof.H) . using again H¨lder’s inequalo ity. ω). 1 0 Thus limε→0 Fε (t) = F (t).
.. ∀ η ∈ C0 .1 Some properties of C0 ∞ First we notice that...t2 . (3. So.t2 ...tn . + η1 − η2 k ) is a complete metric space. . Moreover.Brownian motion 37 3. ω) We come back to the mapping B and we denote by Q its law (which is a probability measure on (C0 . endowed with the metric. ω → B(·.5.. B(C0 )). as easily checked.. the σalgebra generated by C coincides with B(C0 ) since any ball (with respect to the metric of C0 ) is a countable intersection of cylindrical sets. It is important to notice that B(C0 ) is generated by the cylindrical subsets of C0 that we shall introduce now. Note that Ct1 . for any nonnegative Borel mapping F : C0 → R. Q is called the Wiener measure on (C0 . k]}... For n ∈ N.A = Ct1 .16) Some examples of mappings F are the following.. we have E[F (B(·))] = Ω F (B(·.. η2 ) := k=1 2k (1 η1 − η2 k . 3. η k = sup{η(t) : t ∈ [0.tn . Let us now consider the σalgebra B(C0 ). 0 < t1 < · · · < tn and A ∈ B(Rn ) we deﬁne Ct1 . C0 .A := {η ∈ C0 : (η(t1 ). B(C0 )).. η → F (η).t2 . ω))P(dω) = C0 F (η)Q(dη).. η(tn )) ∈ A} .5. d(η1 . Using this identity one can easily see that C is an algebra..tn ..tn+1 .2 The Wiener measure and the standard Brownian motion B : Ω → C0 .. k. n ∈ N.A×Rk .. .. We have set for any k ∈ N.tn+k .
we have Q(Ct1 . B(tn )) ∈ A). where G : Rn → R is nonnegative Borel and t1 .. thanks to (3. t ≥ 0.ω)−B(s. Q) setting W (t)(η) = η(t). called the standard Brownian motion.ω))h P(dω) = E[ei(B(t)−B(s)) ] = e− 2 (t−s)h . Proposition 3. for all η ∈ C0 . W (t) − W (s) is a Gaussian random variable Nt−s . Q)..t2 ... η(tn )).. t ≥ 0.t2 .tn ...A ) = 1 (2π)n t1 (t2 − t1 ) · · · (tn − tn−1 ) A e 1 2 − 2t − 2(t 1 ξ2 (ξn −ξn−1 )2 (ξ −ξ1 )2 −···− 2(t −t n 2 −t1 ) n−1 ) dξ.. For this it is enough to show that the Fourier transform of W (t) − W (s) ψ(h) := C0 ei(η(t)−η(s))h Q(dη).t2 .38 Chapter 3 (i) F (η) = g(η(t0 )). t ≥ 0.16).. B(C0 ). In fact by (3... tn > 0 are given. is given by e− 2 (t−s)h .A ) = P((B(t1 )...A be a cylindrical set. η ∈ C0 . We simply note that.18 Let Ct1 . h ∈ R. where g : R → R is nonnegative Borel and t0 > 0 is given. Proposition 3. Then we have Q(Ct1 . (iii) F (η) = supt∈[0. . The proof is straightforward. In an analogous way one can prove that W (t). Proof.. . Now we deﬁne a stochastic process W (t)...tn .17 W is a Brownian motion in (C0 .16) we have ei(η(t)−η(s))h Q(dη) = C0 Ω 1 2 ei(B(t. so that the conclusion follows from Proposition 3.tn . Let us show for instance that for t > s ≥ 0. has independent increments. for all η ∈ C0 ... (ii) F (η) = G(η(t1 ).. . Let us compute the Wiener measure of a cylindrical set. h ∈ R. for all η ∈ C0 .4. 1 2 h ∈ R.. . Proof. B(C0 ). in (C0 .1] η(t)..
T ] σ = {0 = t0 < t1 < · · · < tn = T }. T ].17) Moreover n 2 EJσ 2 = E k=1 n B(tk ) − B(tk−1 )2 n =E k=1 B(tk ) − B(tk−1 ) + 2 h<k=1 4 EB(th ) − B(th−1 )2 B(tk ) − B(tk−1 )2 . on a probability space (Ω. Proof. For any σ = {0 = t0 < t1 < · · · < tn = T } ∈ Σ(0. F . Then for any σ = {0 = t0 < t1 < · · · < tn = T } ∈ Σ(0. For any T > 0 we denote by Σ(0. F . We introduce a partial ordering on Σ(0. 2 2 E(Jσ − T 2 ) = E(Jσ ) − 2T E(Jσ ) + T 2 = E(Jσ ) − T 2 . setting σ1 ≤ σ2 if and only if σ1  ≤ σ2 . we have E(Jσ ) = T. Since Btk −Btk−1 is a real Gaussian random variable with law Ntk −tk−1 .n − 1}. Then we prove Theorem 3. Let us now introduce the quadratic variation of Brownian motion B in [0. . .Brownian motion 39 3.19 We have σ→0 lim Jσ = T in L2 (Ω. t ≥ 0. P). T ). T ) the set of all decompositions of [0. and so. T ) we set σ := min{tk − tk−1 : k = 1. T ]. (3. T ) we deﬁne n Jσ := k=1 B(tk ) − B(tk−1 )2 . We say that T is the quadratic variation of B in [0. P)..6 Quadratic variation of the Brownian motion In this section we are given a real continuous Brownian motion B(t)..
18) and. Now. (2) .20) on (3. Proposition 3. In fact the following result holds. Then for any σ = {0 = t0 < t1 < · · · < . since B(th ) − B(th−1 ) and B(tk ) − B(tk−1 ) are independent.40 But we have n n Chapter 3 E k=1 B(tk ) − B(tk−1 )4 = 3 k=1 (tk − tk−1 )2 . substituting (3. h<k=1 (3. (3. In other terms the set VT := {ω ∈ Ω : B(·. V (f ) is called the variation of f .19) Therefore n n EJσ 2 = 3 k=1 n (tk − tk−1 )2 + 2 n (th − th−1 )(tk − tk−1 ) h<k=1 2 = 2 k=1 n (tk − tk−1 ) + k=1 2 (tk − tk−1 ) . T )} has outer probability zero. T ] → R of ﬁnite variation.20 We have P∗ (VT ) = 0. BV (0.. ω) ∈ BV (0. (3.17).. we obtain n E Jσ − T  as σ → 0. tn = T } ∈ Σ(0. Let f : [0. An important consequence of Theorem 3.. T ] → R.19 is that almost all trajectories of the Brownian motion B have not bounded variation (2) . 2 =2 k=1 (tk − tk−1 )2 → 0.20) = 2 k=1 (tk − tk−1 )2 + T 2 . T ) we n set Vσ (f ) = k=1 f (tk ) − f (tk−1 ) and deﬁne V (f ) := supσ∈Σ Vσ (f ). T ) is the set of all functions f : [0. we have n n EB(th ) − B(th−1 ) B(tk ) − B(tk−1 ) = h<k=1 2 2 (th − th−1 )(tk − tk−1 ).
. t − s < δε =⇒ B(t.. Then set Bi (t) = Wei 1l[0.. n..t] . tn ∈ [0. Bn are independent Brownian motions. . +∞) the random variables Xi (ti ) are independent.22 Let us construct an ndimensional Brownian motion.Brownian motion Proof.. P) there exists a sequence (σn ) ⊂ Σ(0. . Let (e1 .. Set Λ := {ω ∈ Ω : B(·.. Since ε is arbitrary ω cannot belong to Λ1 . n→∞ We claim that VT ∩ Λ ⊂ Λc . i = 1.. where Q is any operator in L+ (H) such that Ker 1 Q = {0}. Since limσ→0 Jσ = T in L2 (Ω. ω) − B(s. 3. Since B(·.7 Multidimensional Brownian motions Deﬁnition 3. Then one can check easily that B(t) = (B1 (t). F . Example 3. s ∈ [0... for any ε > 0 there exists δε > 0 such that t.. if n is so large that σn  < δε we have Jσn (ω) ≤ εV (B(·. ω)). X1 . T ]. en ) be the canonical basis in Rn and choose Ω = H = L2 (0. Consequently.... .. Then X(t) := (X1 (t). Xn are said to be independent if for any t1 . . 1 (3. Rn ).. F .. F = B(H) and P = NQ . Let ω ∈ VT ∩ Λ. ω) < ε. 1 Let us prove the claim. T ].. P). is called an ndimensional stochastic process. . ω) is uniformly continuous in [0. t ≥ 0.. Bn (t)).. 41 so that P(Λ) = 1 because B is continuous. Bn (t)) is an ndimensional Brownian motion.... .. +∞. .. ∀ t ≥ 0. . A ndimensional Brownian motion is a ndimensional stochastic process B(t) := (B1 (t).. such that B1 .21 Let n ∈ N and let X1 .. T ) such that σn  → 0 and a set Λ1 ⊂ F such that (i) P(Λ1 ) = 1. (ii) lim Jσn (ω) = T for all ω ∈ Λ1 .21) By the claim the conclusion will follow since P(Λc ) = 0. . t ≥ 0. ω) is continuous }. . The claim is proved. Xn be stochastic processes on a probability space (Ω. Xn (t)).
(3. We have n (3. B(t) − B(s) is a Gaussian random variable with law N(t−s)In . Let us check (iii). where In represents the identity in Rn . t ≥ 0. t ≥ 0.23) Z(t) = etA x + 0 e(t−s)A CdB(s).23 Prove that for 0 ≤ s < t we have E B(t) − B(s)4 = (2n + n2 )(t − s)2 . (iii) We have E B(t) − B(s)2 = n(t − s).24) esA CC ∗ esA ds. (ii) E[Bi (t)Bj (t)] = 0 if i = j.24 Let A. Exercise 3.22) E B(t) − B(s) 2 = k=1 E Bk (t) − Bk (s)2 = n(t − s).Qt . (i) If t > s. Then the following properties are easily checked. Exercise 3. where Qt = 0 t ∗ (3. C ∈ L(Rd ) and set t (3. Prove that the law of Z(t) in Rd is given by NetA x.42 Chapter 3 Let B be a Brownian motion in Rn .25) where A∗ and C ∗ are the adjoint of A and C respectively. .
t)} . +∞) → R introduced in Chapter 3 and Q is the Wiener measure. Obviously F0 = {∅. Moreover. < tn . tn ≤ t and A ∈ B(Rn ). t ≥ 0.. Ω}.. let W (t)..1 Filtration Ct1 . stopping time and transition semigroup.tn . The family of σ–algebras (Ft )t≥0 is increasing. Q) where C0 is the complete metric space of all continuous functions ω : [0.A = {ω ∈ C0 : (ω(t1 ). 4.. Moreover. the standard Brownian motion in (C0 . To this purpose we shall introduce some basic concepts as ﬁltration. in particular the Markov and strong Markov property and the reﬂexion principle. W (tn )) ∈ A} For any t > 0 we denote by Ct the algebra of all cylindrical sets where 0 ≤ t1 < . This chapter is devoted to some sharp properties of the Brownian motion.. ω(tn )) ∈ A} = {ω ∈ C0 : (W (t1 ). For any t > 0 we deﬁne Ft− = σ{Ft− : 43 ∈ (0. ∀ t ≥ 0.Chapter 4 Markov property of the Brownian motion Let us consider the probability space (C0 . it is called the natural ﬁltration of W . . we denote by Ft the σalgebra generated by Ct .··· . ω ∈ C0 . Q) deﬁned by W (t)(ω) = ω(t). B(C0 ). B(C0 ).. ...
so that F0+ = F0 . To prove the converse inclusion it is enough to show that Ct ⊂ Ft− .··· .1.tt− 1 . Let in fact I = Ct1 . . Remark 4. k→∞ k so that I ∈ Ft− as well. Proposition 4. It is clear that Ft ⊃ ∈(0. Notice that A = {ω ∈ Ω : ω (0) = 0}. Proof. so that Ft ⊃ Ft− . An ∈ F0+ .tn . Then An ∈ F1/n and A = n∈N n ∈ N. t) and Ft+ : = >0 Ft+ .1 Ft measurable random variables I ∈ B(R) ⇒ X −1 (I) ∈ Ft . If tn < t then I belongs to Ft− whereas if tn = t we have I = lim Ct1 . t ≥ 0. The following lemma will be frequently used.··· . Due to Proposition 4.t) Ft− .1 For all t > 0 we have Ft = Ft− .44 where σ ∈(0. We say that a real random variable X is Ft measurable if In this case we say also that X depends from the story of the Brownian motion only up to t.A ∈ Ct so that tn ≤ t. 4. Let t > 0.t) Chapter 4 Ft− is the σalgebra generated by Ft− for ∈ (0. Let for instance t = 0 and consider the sets An = {ω ∈ Ω : ω(1/n) ≤ 1/n}.1 we say that the natural ﬁltration (Ft )t≥0 is left continuous. that is Ft+ = Ft for all t ≥ 0.2 The ﬁltration (Ft )t≥0 is not right continuous.A ∈ Ft− .
It is enough to show that for any A ∈ Ft .4 (onezero law) Assume that A ∈ F0+ .I = {ω ∈ Ω : (ω(t1 + h) − ω(h). . (4. and let ϕ be a real random variable Ft –measurable. In fact if A ∈ D it is obvious that Ac ∈ D. I ∈ B(Rn ).. D = {A ∈ Ft : 1 A is independent of W (s2 ) − W (s1 )}. Denote by G the σalgebra generated by all sets of the form Dt1 . Remark 4.1) On the other hand. Proof. we claim that G = B(C0 ). Then W (s2 ) − W (s1 ) and ϕ are independent. D is a λsystem..tn − 1 . D contains the algebra of all cylindrical set belonging to Ct (which is a πsystem). so that P2 (A) = P(A) which yields P(A) equal to zero or one.1 in Appendix A).4 one can easily show that (Ft )t≥0 is both right and left continuous.I belongs to G .. . where n ∈ N.tn .tn . Moreover..1) G = A.Markov property 45 Lemma 4. W (s2 ) − W (s1 ) and 1 A are l independent....tn . .I . Moreover. Then either P(A) = 1 or P(A) = 0.. Proposition 4. and W has independent increments. ω(tn + h) − ω(h)) ∈ I}. Next result shows that F0+ contains only trivial sets...... l Since W is a process with independent increments. if (An ) is a sequence in D consisting of disjoint sets. but this follows from the identity j→∞ lim Dt1 − 1 . ω(tn ) − ω(1/j)) ∈ I} = Ct1 . By using Proposition 4. t > 0.. since it belongs to all Ft .. To prove the claim it is enough to show that any cylindrical set Ct1 .3 Let s2 > s1 ≥ t > 0.. in other words that Ft coincides with the set D deﬁned below. Then we have P(A ∩ G) = P(A)P(G). It is clear that A is independent of G ... j→∞ Since G = B(C0 ) we can set in (4.5 For any t ≥ 0 denote by Ft the σalgebra generated by Ft and all null sets of Ω (called the completion of Ft ).h... h > 0. 1 . Let A ∈ F0+ . 0 < t1 < · · · < tn .h. Now the claim follows from Dynkin’s theorem (Theorem A. one can show easily that ∞ n=1 An ∈ D. ∀ G ∈ G.. Proof.I j j j = lim {ω ∈ Ω : (ω(t1 ) − ω(1/j)..
46 Chapter 4 4. ω(tn )) ∈ I} = Ct1 .tn . In fact (τ ) Ct1 ..I = {ω ∈ Ω : tn (ω) < τ.tn ... . Let us describe the σalgebra Fτ .. Q) is called a stopping time with respect to the ﬁltration (Ft )t≥0 if {τ ≤ t} ∈ Ft for all t ≥ 0.I in included in Fτ and one can show that it coincides with Fτ . +∞]) random variable τ in (C0 . Moreover. We claim that Ct1 .. but it is a stopping time with respect to the ﬁltration (Ft+ )t≥0 ....I ∩ {τ ≤ t} = Ct1 . B(C0 )......6 Let τ be an extended random variable such that {τ < t} ∈ Ft ... τ is Fτ measurable. if A = {τ ≤ s} we have A ∩ {τ ≤ t} = {τ ≤ t ∧ s} ∈ Ft∧s ⊂ Ft . In fact ∞ {τ ≤ t} = k=1 τ ≤t+ 1 k ∈ Ft+ .. (τ ) (τ ) (τ ) Then τ is not in general a stopping time with respect to (Ft )t≥0 .I ∩ {tn < τ ≤ t} So.tn . the σalgebra generated by all Ct1 .I ∩{tn < τ }.tn ...tn . for all t ≥ 0.. In fact. < tn and I∈B(R) we deﬁne Ct1 .. For 0 < t1 < ..2 Stopping times A nonnegative extended (that is with values in [0...tn .I is Fτ measurable. In other words we have Fτ ⊃ σ(τ ).. .. (ω(t1 ). Remark 4.. If τ is stopping time.. where σ(τ ) is the σalgebra generated by τ . To any stopping time τ we associate the σalgebra Fτ : = {A ∈ F : A ∩ {τ ≤ t} ∈ Ft for all t ≥ 0}... then {τ > t} and {τ = t} belong obviously to Ft for all t ≥ 0.
that is A ∩ {τ ≤ t} ∈ Ft . (4. . n 2 k k−1 ≤τ < n n 2 2 ∈ Ft . Proposition 4. In fact. ∀ k ∈ N. k 2n =A∩ k−1 k ≤τ < n n 2 2 ∈Fk. Deﬁne for any n ∈ N and ω ∈ Ω τn (ω) = k 2n if k−1 k ≤ τ (ω) < n . Then we have A ∩ τn = so that A ∈ Fτn . We want to extend several properties concerning time t to general stopping times τ . if t = 2k with k ∈ N we have n {τn = t} = Finally.2) It is clear that the sequence (τn ) is decreasing.9 Let τ be a stopping time and set Wτ (ω) = W (τ (ω). We start by showing that Wτ is Fτ measurable. (4. Assume ﬁrst τ discrete.8 Let τ be a stopping time. n 2 2 k ∈ N. 0 < t1 < · · · < tk < · · · ω ∈ Ω. ω). let A ∈ Fτ . Show that τ is a stopping time if and only if {τ = µk } ∈ Fµk for all k ∈ N.3) ∀ t ≥ 0. τ (Ω) = {tk }. Proposition 4. Then there exists a decreasing sequence (τn ) of discrete stopping times convergent pointwise to τ such that Fτn ⊃ Fτ for all n ∈ N. that is that τ (Ω) = (µk )k∈N where µk is an increasing sequence of positive numbers. Then Wτ is Fτ measurable. Proof. Show that in this case Fτ is the σ–algebra Fτ : = {A ∈ F : A ∩ {τ = µk } ∈ Fµk for all k ∈ N}. Proof. τn is a stopping time.7 Assume that the nonnegative random variable τ is discrete. Moreover.Markov property 47 Exercise 4.
By the previous argument we have {Wτn ∈ I} ∩ {τn ≤ t} ∈ Ft Now the conclusion follows letting n → ∞. τ is a stopping time with respect to ﬁltration {Ft+ }t≥0 .10 Let a ∈ R and set (1) for all I ∈ B(R). k ∈ N.t] {W (s) ≤ a} = {W (s) ≤ a} ∈ Ft .t]∩Q Consequently. ω ∈ Ω.6. ω). . s∈[0. lim Wτn (ω) = Wτ (ω). Then we have Wτ (ω) = W (tk )(ω). Then we have {τ ≥ t} = s∈[0. Let now τ be arbitrary. (4. Since W is continuous we have n→∞ ω ∈ Ω. Let I ∈ B(R). Example 4.t] {W (s) < a} = {W (s) < a} ∈ Ft . let τn be deﬁned by (4.4) τa = inf{t ≥ 0 : W (t) = a}. ∈ I} ∩ {τ ≤ t} ∩ Ak ] ∈ I} ∩ {τ ≤ t} ∩ Ak ] ∈ I} ∩ {τ ≤ t} ∩ Ak ] ∈ Ft . by Remark 4. Let now τ = inf{t ≥ 0 : W (t) > a}.48 and set Ak = {τ = tk }. the conclusion holds in this case. k ∈ N.2) and set Wτn (ω) = W (τn (ω). τa is a stopping time with respect to the ﬁltration (Ft )t≥0 .t]∩Q So. Then {τa > t} = s∈[0. Fix t ≥ 0. (1) We use the convention that the inﬁmum of the empty set is +∞. ∞ {k∈N: tk ≤t} [{Wtk So. s∈[0. Then {Wτ ∈ I} ∩ {τ ≤ t} = = = ∞ k=1 [{Wtk ∞ k=1 [{Wτ Chapter 4 ∀ω ∈ Ak .
E eiα(W (t+τ )−W (τ )) = i=1 Ai eiα(W (t+ti )−W (ti )) dP = i=1 E 1 Ai eiα(W (t+ti )−W (ti )) . it follows that l ∞ E e iα(W (t+τ )−W (τ )) = i=1 P(Ai )E eiα(W (t+ti )−W (ti )) = e− 2 α 1 2t and so (4. . 1 2 α ∈ R. Continuity of C(t) is obvious. Let now τ be general and let (τn ) be the sequence of ﬁnite stoppping times deﬁned by (4. By (4.2). Proposition 4. Proceeding similarly one can prove that the law of C(t) − C(s) with t > s > 0 is Nt−s and that C(t) has independent increments. We have just proved that E eiα(W (t+τn )−W (τn )) = e− 2 α t .5) it follows that C(t) is a Gaussian random variable Nt . τ (Ω) = (tk ) and set Ai = {τ = ti } ∈ Fti . t ≥ 0. Let us ﬁrst prove that the law of C(t) is Nt .11 Let τ be a stopping time. Proof. (4. Assume ﬁrst that τ is discrete. α ∈ R. Now (4. is a Brownian motion. We want now to show that the same holds when h is replaced by a stopping time.Markov property 49 4.5) is proved. is a Brownian motion for any h > 0.5) ∀ i ∈ N.3 The Brownian motion W (t + τ ) − W (τ ) We recall that W (t + h) − W (t).5) follows letting n tend to inﬁnity. l Since 1 Ai and W (t + ti ) − W (ti ) are independent. Then C(t) := W (t + τ ) − W (τ ). For this it is enough to show that for any α ∈ R we have E eiαC(t) = E eiα(W (t+τ )−W (τ )) = e− 2 α t . Then we have ∞ ∞ 1 2 t ≥ 0.
that is P0 = I and where gt (ξ) = √ Pt+s = Pt Ps . we deﬁne the transition semigroup Pt ϕ(x) = E[ϕ(W (t) + x)]. ∀ t.50 Chapter 4 4.6) Since the law of W (t) + x is Nx. t ≥ 0.12 One can show that u(t. is the unique solution of the Dirichlet problem above. ∀ t > 0. o Exercise 4.13 Prove that for t > s ≥ 0. There is a simple deterministic proof based on maximum principle and a stochastic proof. 2 u(0. x ∈ R. ξ2 1 e− 2t .4 Transition semigroup We shall denote by Bb (R) the set of all real. +∞) × R → R. t ≥ 0. x). by an explicit computation. inﬁnitely diﬀerentiable and fulﬁlls ut (t. u(t. x ∈ R. t > 0. x) = Pt ϕ(x) is continuous. Pt−s ϕ(x) = E[ϕ(W (t) − W (s) + x)]. x ∈ R.7) = −∞ gt (x − y)ϕ(y)dy. Notice that Pt coincides with the heat semigroup in R. ϕ ∈ Bb (R). (4. (4. is a semigroup of linear operators in Bb (R). ∀ x ∈ R. Given ϕ ∈ Bb (R) we want to study the evolution in time of ϕ(W (t) + x). bounded and Borel functions and by Cb (R) the subspace of Bb (R) of those functions which are uniformly continuous and bounded on R. x ∈ R.8) 2πt We deduce. (4. which we will present later. To this purpose. In fact one checks easily that if ϕ ∈ Cb (R) then the function u : [0. s ≥ 0.t we have Pt ϕ(x) = E[ϕ(W (t) + x)] = √ 1 2πt +∞ +∞ −∞ e− 2t (x−y) ϕ(y)dy 1 2 (4.9) . x) = ϕ(x). x) = Pt ϕ(x). Remark 4. ϕ ∈ Bb (H). based on Itˆ’s formula. x) = 1 uxx (t. ξ ∈ R. t ≥ 0. that Pt .
where (recall Exercise 4. Notice that U is Fs measurable and V is independent of Fs . We are here concerned with the stochastic process X(t) = X(t. (4. .5 Markov property In this section we shall use several properties of conditional expectation. η a Fs measurable random variable and ϕ ∈ Bb (R).11) Moreover X(·) is a Markov process.10) (Pt−s ϕ)(X(s))dP. Exercise 4.10) is proved. Proof. they are recalled in Appendix A.15 Let s > 0. Equivalently ϕ(X(t))dP = A A t ≥ 0. So. To prove the last statement notice that by Proposition B.3 we have E[ϕ(X(t))X(s)] = E [E[ϕ(X(t))Fs ]X(s)] = E[Pt−s ϕ(X(s))X(s)] = Pt−s ϕ(X(s)) = E[ϕ(X(t))Fs ]. (4.13) h(u) = E[ϕ(u + V )] = E[ϕ(u + W (t) − W (s))] = Pt−s ϕ(u). By Proposition B. Show that E[ϕ(W (t) + ηFs ] = (Pt−s ϕ(η)). where x ∈ R.Markov property 51 4. Set X(t) = W (t) + x = (W (s) + x) + (W (t) − W (s)) =: U + V. Proposition 4.14 For any t > s > 0 and any ϕ ∈ Bb (H) we have E[ϕ(X(t))Fs ] = (Pt−s ϕ)(X(s)).6 it follows that E[ϕ(X(t))Fs ] = E[ϕ(U + V )Fs ] = h(U ). x) = W (t) + x. (4. ∀ A ∈ Fs .
.Then we have E[ϕ(X(t))Fτ ] = (Pt−τ ϕ)(X(τ )). Let A ∈ Fτ ..5. Then we have ∞ (Pt−τ ϕ)(W (τ ))dP = A i=1 A∩{τ =ti } (Pt−τ ϕ)(W (τ ))dP ∞ = i=1 A∩{τ =ti } (Pt−ti ϕ)(W (ti ))dP.. we can write. (4. (4. Therefore. Proposition 4. by (4.16 Let τ be a stopping time and let t ≥ τ and ϕ ∈ Bb (H). . i = 1.10) and taking into account that by the deﬁnition of Fτ we have A ∩ {τ = ti } ∈ Fti .1 Strong Markov property We now consider conditional expectation with respect to Fτ where τ is a stopping time. We set x = 0 for simplicity. Therefore. (4.13) Proof.12) Equivalently ϕ(X(t))dP = A A (Pt−τ ϕ)(X(τ ))dP. Assume ﬁrst that τ is of the form τ (Ω) = (tk )k∈N . so that X(t) = W (t). ∀ A ∈ Fτ .52 Chapter 4 4. n.13) is proved. . ∞ (Pt−τ ϕ)(W (τ ))dP = A ∞ i=1 A∩{τ =ti } (Pt−ti ϕ)(W (ti ))dP = i=1 ∞ A∩{τ =ti } E[ϕ(W (t))Fti ]dP = i=1 A∩{τ =ti } ϕ(W (t))dP = A ϕ(W (t))dP.
taking into account that {Ta ≤ t} = {M (t) ≥ a} (4. We have.13) it follows that ϕ(W (t))dP = A A (Pt−τn ϕ)(W (τn ))dP for all A ∈ Fτ . • Tb = inf{t ≥ 0 : B(t) = b}.14) . s∈[0.15) To ﬁnd the laws of Ta with a ≥ 0 and M (t) the following lemma is useful.12) is called the strong Markov property of W . Let A ∈ Fτ .8) F τ ⊂ F τn for all n ∈ N.Markov property 53 Let now τ be an arbitrary stopping time and let (τn ) be deﬁned by (4. a ≥ 0 (4. • M (t) = max B(s). t ≥ 0. M (t) ≥ a) = P(B(t) ≥ a).6 Some consequences of the strong Markov property In this section we want to determine the laws of the following important random variables. s∈[0.16) t ≥ 0. Then we have P(B(t) ≤ a. • m(t) = min B(s). Lemma 4. t ≥ 0. and {Ta ≤ t} = {m(t) ≤ a}.t] b ∈ R. Property (4.17 Let a ≥ 0 and t ≥ 0. (4. Recall that (Proposition 4. Now the conclusion follows letting n → ∞. t ≥ 0.t] Notice that {Ta ≤ t} = {M (t) ≥ a}. Proof. Then by (4. a ≤ 0. 4.2).
18 (Reﬂection principle) For all a ≥ 0 we have P(M (t) ≥ a) = 2P(W (t) ≥ a). M (t) ≥ a) = {Ta ≤t} E[1 (−∞. (4.+∞) (a)]dP l = {Ta ≤t} E[1 [a. Proof. W (t) ≥ a). M (t) ≥ a) = P(W (t) ≥ a). a > 0. M (t) ≥ a) = {Ta ≤t} ∀ s > 0.17) .a] (a)]dP.a] (W (t))FTa ]dP. Ps 1 (−∞. l On the other hand. as easily checked. l E[Pt−Ta 1 (−∞. l since {Ta ≤ t} ∈ FTa .a] (a)]dP = {Ta ≤t} E[Pt−Ta 1 [a.+∞) (a). Proposition 4.a] (W (t))FTa ]dP l = {Ta ≤t} E[Pt−Ta 1 (−∞. Ta ≤ t) = {Ta ≤t} Chapter 4 1 (−∞. W (t) ≤ a) + P(M (t) ≥ a.a] (W (Ta ))]dP l = {Ta ≤t} E[Pt−Ta 1 (−∞.a] (a) = Ps 1 [a. we have.54 P(W (t) ≤ a.a] (W (t))dP l = {Ta ≤t} E[1 (−∞.+∞) (W (t))FTa ]dP l = P(W (t) ≥ a. Write P(M (t) ≥ a) = P(M (t) ≥ a. M (t) ≥ a) = P(W (t) ≤ a. By the strong Markov property it follows that P(W (t) ≤ a. l l Therefore P(W (t) ≤ a.
. Corollary 4. Moreover.19 it follows that at ﬁxed time t the law of M (t) coincides with that of W (t). Then we have a2 a ((Ta )# P)(dt) = √ e− 2t dt.20 From Corollary 4. 2πt (4.19 (Law of M (t)) For all t ≥ 0 we have ξ2 2 (M (t)# P)(dξ) = √ l e− 2t 1 [0. η2 a2 d a P(Ta ≤ t) = √ e− 2t dt. the conclusion follows.+∞) (ξ)dξ.18 for any a ≥ 0 P(M (t) ≥ a) = 2P(W (t) ≥ a) = √ 2 2πt +∞ a e− 2t dξ ξ2 = P(W (t) ≥ a). by Lemma 4. The following results can be proved similarly.18 we have 2 P(Ta ≤ t) = P(M (t) ≥ a) = √ 2πt 2 =√ 2π Therefore +∞ at−1/2 +∞ a e− 2t dξ ξ2 e− 2 dξ. By Proposition 4. +∞)) are diﬀerent.19) Proof.14) and Proposition 4. W (t) ≥ a) = P(W (t) ≥ a) so.18 we can easily deduce the expressions of the laws of M (t) and Ta for all a ∈ R. though random variables M (t) and W (t) are diﬀerent.21 (Law of Ta ) Let a ≥ 0 and t ≥ 0. Remark 4.17 we have P(M (t) ≥ a.Markov property 55 Now. Corollary 4. it is clear that P(M (t) ≥ a. We have in fact by Proposition 4. dt 2πt3 which implies the conclusion. Obviously the laws of M (·) and W (·) on C0 ([0. W (t) ≤ a) = P(W (t) ≥ a).18) Proof. in particular M (t) is increasing whereas W (t) is not. 2πt3 (4. By (4.
a] (ξ)dξ.21) Corollary 4. (4.20) Proposition 4. ((Ta )# P)(dt) = √ 2πt3 (4. absorbed in 0 t ≥ 0.22 Let a ≤ 0 and t ≥ 0.23 (Reﬂection principle) For all a ≤ 0 we have P(m(t) ≤ a) = 2P(W (t) ≤ a).56 Lemma 4.7 Application to partial diﬀerential equations τx = inf{t ≥ 0 : W (t) + x = 0} = T−x . (ii) U (t) = W (t) + x. U (t) is called the Brownian motion reﬂected in 0 (iii) V (t) = W (t ∧ τx ) + x. Then we have a2 a e− 2t dt. 2πt (4. m(t) ≤ a) = P(W (t) ≤ a). Then we have P(W (t) ≥ a. x ≥ 0.23) 4. For any x ≥ 0 we set in this section Moreover we consider the following processes which take values in [0. τx ]. +∞). Y (t) is called the Brownian motion killed in 0. (i) Y (t) = W (t) + x.24 (Law of m(t)) For all t ≥ 0 we have (m(t)# P)(dξ) = − √ ξ2 2 e− 2t 1(−∞.25 (Law of Ta ) Let a ∈ R and t ≥ 0. Chapter 4 (4. ∀ t ∈ [0. V (t) is called the Brownian motion . t ≥ 0.22) Corollary 4.
x ≥ 0. Proposition 4. ut (t. l l where ψ(λ) = 1 t>λ l 1 2π(t − λ) R x ≥ 0. x ≥ 0.8). Write E[ϕ(W (t) + x)1 t>τx ] = E[E[1 t>τx ϕ(W (t) + x)Fτx ]] l l = E[1 t>τx E[ϕ(W (t) + x)Fτx ]] l Now. x ∈ H.26 We have +∞ u(t. Deﬁne for any ϕ ∈ Bb ([0. 0) = 0.1 The Dirichlet problem in the halfline ∀ t ∈ [0. (4. E[ϕ(W (t) + x)1 t>τx ] = E[1 t>τx (Pt−τx ϕ)(0)] =: E[ψ(τx )]. l where ϕ is extended to R by setting ϕ(−x) = ϕ(x). (4. We have u(t. x). t > 0. l t ≥ 0. u(0. . x) is the solution of the Dirichlet problem in [0.7. t > 0 2 (4.Markov property 57 4. +∞)) Ut ϕ(x) := u(t. Proof. x) = ϕ(x). +∞). using the strong Markov property we ﬁnd that. τx ]. t ≥ 0. We are going to show that u(t. x > 0. x) := E[ϕ(W (t) + x)1 t≤τx ]. e− 2(t−λ) ϕ(ξ)dξ. x) = 0 [gt (x − y) − gt (x + y)]ϕ(y)dy. x) = 1 uxx (t.24) We are here concerned with the process Y (t) = W (t) + x. x) = E[ϕ(W (t) + x)1 t≤τx ] l = Pt ϕ(x) − E[ϕ(W (t) + x)1 t>τx ].26) where g is deﬁned by (4. ξ2 λ > 0.25) u(t.
a We recall that Erfc (a) = 2 √ π . (x+y)2 1 ∂ Gx.y ϕ(y)dy. by a direct computation.23)) it follows that t Chapter 4 E[ϕ(W (t) + x)1 t>τx ] = l 0 R t gt−s (y)ϕ(y)dy √ ∂ ∂x x 2πs3 e− 2s ds x2 = gt−s (y)ϕ(y)dy gs (x)ds 0 R = R gt (x − y)ϕ(y)dy + ∂ ∂x Gx. Ut ϕ(x) = u(t. x) is the solution of the Dirichlet problem (4. R where (2) t Gx. 4. We consider the process For any ϕ ∈ Bb ([0. x) = R gt (x − y)ϕ(y)dy − R gt (x + y)ϕ(y)dy.58 Next. Since.y = − √ e− 2t = −gt (x + y) ∂x 2πt we get u(t. s ≥ 0.2 The Neumann problem U (t) = W (t) + x. It is easy to check.7. x ≥ 0. Replacing in the last integral y with −y.y = 0 gt−s (y)gs (x)ds = 1 Erfc 2 x + y √ 2t . that if ϕ ∈ Cb ([0. and the conclusion follows. +∞)). t ≥ 0. for x > 0. Moreover U0 = I and Ut+s = U (t)U (s) for all t. we see that +∞ Qt ϕ(x) = 0 (2) [gt (x − y) + gt (x + y)]ϕ(y). +∞ −r 2 e dr.25). recalling the law of τx (see (4. +∞)) we set Qt ϕ(x) = E[ϕ(W (t) + x)] = (2πt)−1/2 R e− x−y2 2t ϕ(y)dy.
Zt ϕ(x) = Ω ϕ ∈ Bb ([0. x) = Zt ϕ(x) we see that u is the solution to the Ventzell problem. x). s ≥ 0. ∞) and solves the following Neumann problem ut (t. 4. x) = ϕ(x). . 2 ux (t. t ≥ 0 2 +∞ Zt ϕ(x) = uxx (t. {t≥τx } = {t<τx } since W (τx ) + x = 0.Markov property 59 where gt is deﬁned by (4. x ≥ 0. inﬁnitely diﬀerentiable in (0. [gt (x − y) − gt (x + y)]ϕ(y)dy + √ 2πt −∞ 0 If ϕ ∈ Cb ([0. where Ut is deﬁned by (4. 0) = 0. x). x) = 1 uxx (t.24). So x y2 ϕ(0) e− 2t dy. ∞). ∞) × [0. +∞)) then u(t. x ≥ 0. where x ≥ 0.3 The Ventzell problem V (t) = W (t ∧ τx ) + x. +∞)). Therefore Zt ϕ(x) = Ut ϕ(x) + ϕ(0) P(T−x ≤ t). ϕ(B(t ∧ τx ) + x)dP ϕ(W (t) + x)dP + ϕ(0)dP. setting u(t. t > 0. So. Moreover Q0 = I and Qt+s = Q(t)Q(s) for all t. Let us consider the stochastic process. 0) = 0. x ≥ 0. x) = ϕ(x).7. x) = 1 uxx (t. x ≥ 0. ut (t.8). Now it is easy to check that if ϕ ∈ Cb ([0. Set Zt ϕ(x) = E[ϕ(W (t ∧ τx ) + x)]. t ≥ 0. u(0. +∞)). ∞) × [0. x ≥ 0. x) = Qt ϕ(x) is continuous in [0. u(0. t ≥ 0. t > 0.
s ≥ 0. Chapter 4 .60 Moreover Z0 = I and Zt+s = Z(t)Z(s) for all t.
1.A = {ω ∈ C0 : (B(t1 ).. l 61 (5. T ].tn .ti ) . we denote by Ft the σalgebra generated by Ct and all Pnull sets of Ω.1 5. We call Ft ..Chapter 5 The Itˆ integral o In all this chapter B represents a Brownian motion in a probability space (Ω..··· . Moreover. Similarly as in Chapter 4. We denote by (Ft )t≥0 the completion of the natural ﬁltration of B with all Pnull sets of Ω. is adapted to the Brownian motion B if F (t) is Ft measurable for any t ∈ [0. in (Ω. tn ≤ t and A ∈ B(Rn ).1) . . P) is a stochastic process of the form n F = i=1 Fi−1 1 [ti−1 .1 Deﬁnition of Itˆ’s integral o Itˆ’s integral for elementary processes o Deﬁnition 5. P). The family of σ–algebras (Ft )t≥0 is increasing.. for any t > 0 we denote by Ct the algebra of all cylindrical sets Ct1 . t ∈ [0. F . F . < tn . t ∈ [0. 5. T ]. B(tn )) ∈ A} where 0 ≤ t1 < . it is called the natural ﬁltration of B.1 Let T > 0. T ]. An elementary process F (t). We say that a stochastic process F (t).. t ≥ 0 the natural ﬁltration of B augmented with the null sets of P.
2) Obviously any elementary process is adapted.2 Assume that F ∈ EB (0. . T ). we deﬁne the Itˆ integral o setting T n I(F ) : = 0 F (s)dB(s) = i=1 Fi−1 (B(ti ) − B(ti−1 ))..3) (5. Let us prove (5. it is independent of B(tj )−B(tj−1 ). We have n E[I(F )2 ] = E j=1 Fj−1 2 [B(tj ) − B(tj−1 )]2 +2E j<k Fj−1 Fk−1 [B(tj ) − B(tj−1 )] [B(tk ) − B(tk−1 )] . .4) E 0 F (s)dB(s) = 0 Proof. (5. Let us prove (5. For any elementary process F (t). n − 1. t ∈ [0. 2 Proposition 5.3) is proved. 1. by Lemma 4. Since Fj−1 is Fj−1 measurable. Therefore we have n E[I(F )] = j=1 E[Fj−1 ]E[B(tj ) − B(tj−1 )] = 0 and (5. We have n E[I(F )] = j=1 E[Fj−1 (B(tj ) − B(tj−1 ))]. 0 = t0 < t1 < · · · < tn = T and Fi is Fti measurable for any i = 0. This property is needed to prove some basic identities (similar to those obtained for the Wiener integral) which allow to extend the integral to more general processes.3).62 The Itˆ integral o where n ∈ N.4). Notice now that for j < k the random variable Fj−1 Fk−1 [B(tj ) − B(tj−1 )]. F . P) and we have T E 0 T F (s)dB(s) 2 T =0 E(F (s)2 )ds. (5.. Then I(F ) ∈ L2 (Ω. T ].3..
·)F1 (t. T ) × F and such that T F ZT := E 0 F (t. T ) × F . Hint: Use the identity ab = 1 1 1 (a + b)2 − a2 − b2 . as required. b ∈ R. F1 = E 0 F (t. 2 Exercise 5. . ω) → F (t. G ∈ EB (0. Prove that T T T E 0 F (s)dB(s) 0 G(s)dB(s) = 0 E[F (s)G(s)]ds. ·)dt. 5. ω). which are measurable with respect to the product σalgebra. (t. Therefore.2 General deﬁnition of Itˆ’s integral o Let us denote by ZT := L2 ([0. It follows that E[I(F ) ] = j=1 2 n E[Fj−1 2 ](tj − tj−1 ). T ).Chapter 5 63 is Fk−1 –measurable and consequently is independent of B(tk ) − B(tk−1 ). Obviously any elementary process F belongs to Z. ·)2 dt < ∞. B(0. The scalar product on Z is deﬁned by T F. taking the expectation. T ] × Ω. T ] × Ω.3 Let F.1. 2 2 2 a. dt × P) the Hilbert space of all (equivalence classes of) functions F : [0. B(0. we have E [Fj−1 Fk−1 [B(tj ) − B(tj−1 )][B(tk ) − B(tk−1 )]] = E [Fj−1 Fj−1 [B(tj ) − B(tj−1 )]] E[B(tk ) − B(tk−1 )] = 0.
Therefore it can be uniquely extended to the closure EB (0. b. Moreover. T ) ⊂ ZT → L2 (Ω. T ] we have c b c F (s)dB(s) = a a F (s)dB(s) + b F (s)dB(s). T ) in ZT .3 it follows that if F and G are predictable square integrable processes we have T T E 0 F (s)G(s)dB(s) = 0 E[F (s)G(s)]ds. the mapping T 2 EB (0. b a (5. T )).64 In view of (5. 2 is an isometry. T E 0 T F (s)dB(s) 2 T =0 E(F (s)2 )ds.6) E 0 F (s)dB(s) = 0 Moreover. So. T ) 2 of EB (0.b) . Note ﬁrst that an elementary process is a linear combination of processes of the form F 1 [a.7) We can deﬁne in an obvious way the Itˆ integral o interval [a. the Itˆ integral can be uniquely deﬁned by extension for any preo dictable square integrable process F (t). FT . t ≥ 0 and the following properties are fulﬁlled.4). for any a. T ]. b] ⊂ [0. l with F Fa measurable. and b b E a F (s)dB(s) = a (EF (s)2 )ds. T ) are called predictable. . We have b F (s)dB(s) in any E a F (s)dB(s) 2 = 0. (5. P)F → 0 The Itˆ integral o F (s)dB(s). c ∈ [0. from Exercise 5.5) (5. 2 Processes belonging to EB (0. Let us now present a characterization of predictable processes (that is of 2 space EB (0.
1). φn → φ = 1lA in L2 ([0. T ]) in L2 ([0. T ] × Ω. dt × P) such that T F (s)dB(s) = 0. P. l Then. b) a predictable rectangle. it is natural to approximate a general predictable process by linear combinations of functions of the form 1 A×[a. Prove that t t ϕ s F (r)dB(r) = s ϕ F (r)dB(r). dt × P). . let us show (A.6 Let F ∈ L2 ([0. 0 Show that F = 0. T ]) is precisely L2 ([0. We call A × [a. P. t] ⊂ [0. Then we set D = {A ∈ P : 1 A ∈ ΛT }. So.1)(iii). T ] × Ω.4 A real predictable process in [0. Fs . For this we shall use the Dynkin Theorem. A ∈ D and (A. Deﬁnition 5. l We claim that D is a λsystem. Exercise 5.5 The closure EB ([0.b) .1)(iii) is fulﬁlled. Properties (B. by the monotone convergence theorem. Since any element of L2 ([0. T ] × Ω. P). dt × P) where A = ∞ Ak . that it fulﬁlls (A.e.1. P. i.1)(i)(ii) are clear. T ] × Ω. T ] × Ω. dt × P) can be approximated by a monotonic sequence of simple functions. P. l with A Fa measurable. So. it is enough to show that 1lA ∈ ΛT for any A ∈ P. see Appendix A. Now k=1 the conclusion follows by Theorem A. 2 Proposition 5.8) Exercise 5. P. dt × P). T ]×Ω. T ] and let ϕ ∈ L∞ (Ω. We denote by R the family of all predictable rectangles and by P the σalgebra generated by R.7 Let F ∈ L2 ([0. T ] × Ω. Let (An ) ⊂ D be mutually disjoint sets and set n φn = k=1 1 Ak . P. P. T ] is a real random variable in the probability space ([0.Chapter 5 65 In turn each F can be approximated by linear combinations of characteristic functions of Fa measurable sets. Denote by ΛT the closure of EB ([0. [s. P is called the σalgebra of all predictable events. dt×P). We ﬁrst note that R is a πsystem. dt × P). (5. 2 Proof.
L2 (Ω)) then F (t) is Ft measurable for all t ∈ [0. (5. For any decomposition σ = {t0 .. tn } ∈ Σ(0. T ] → L2 (Ω.10) Example 5. in L2 ([0. F . 2 (5. P). t1 . t → F (t). P. T ) consider the elementary process n Fσ := j=1 F (tj−1 )1 [tj−1 ..9) σ→0 Consequently we have T σ→0 lim Iσ (F ) = 0 F (s)dB(s) in L2 (Ω. F . 2 Clearly Fσ ∈ EB (0. 2 2 2 . dt × P). T ) and. Write B(tk−1 )(B(tk ) − B(tk−1 )) = B(tk−1 )B(tk ) − B 2 (tk−1 )) 1 1 1 1 = − B 2 (tk ) + B(tk−1 )B(tk ) − B 2 (tk−1 ) + B 2 (tk ) − B 2 (tk−1 ) 2 2 2 2 = 1 2 1 1 B (tk ) − B 2 (tk−1 ) − (B(tk ) − B(tk−1 ))2 . T ] × Ω. T ]. T ] and the mapping [0. . T ].. · · · .tj ) l and set T n Iσ (F ) := 0 Fσ (s)dB(s) = j=1 F (tj−1 )(B(tj ) − B(tj−1 )).66 The Itˆ integral o 5. is continuous. L2 (Ω)) the space of all stochastic processes which are mean square continuous and adapted. We recall that if F ∈ CB ([0.2 Itˆ integral for mean square continuous o processes We shall denote by CB ([0. P).8 Let us prove that T 0 1 B(t)dB(t) = (B 2 (T ) − T ). t1 .11) Let σ = {t0 . tn } ∈ Σ(0. T ). using the continuity of F one can check easily that lim Fσ = F. (5.
has orthogonal increments (in the sense of L2 (Ω. F . However. P). 2 and n σ→0 lim B k=1 tk + tk−1 2 (B(tk ) − B(tk−1 )) = 1 2 B (T ). F .Chapter 5 Then we have Iσ (B) = 1 1 2 B (T ) − 2 2 n 67 (B(tk ) − B(tk−1 ))2 .3 The Itˆ integral as a stochastic process o t Let F ∈ L2 ([0. is not a process with independent increments in general (unless f is deterministic). F . We ﬁrst notice that X(t). P)) as the following result shows. we deduce that T 1 B(t)dB(t) = lim Iσ (B) = (B 2 (T ) − T ). in L2 (Ω. k=1 Recalling that the quadratic variation of B is T (Theorem 3. P). Proposition 5. P. t ∈ [0. X(t). dt × P and set X(t) = 0 F (s)dB(s). Then we have E[(X(t2 ) − X(t1 ))(X(t4 ) − X(t3 ))] = 0 .19). T ] × Ω. σ→0 2 0 Exercise 5.10 Let 0 ≤ t1 ≤ t2 ≤ t3 ≤ t4 ≤ T . t ≥ 0.9 Prove that n σ→0 lim B(tk )(B(tk ) − B(tk−1 )) = k=1 1 (B 2 (T ) + T ). t ≥ 0. in L2 (Ω. 5. 2 t ≥ 0. take for instance t X(t) = 0 B(s)dB(s) = 1 (B 2 (t) − t). T ]. 2 Therefore the deﬁnition of the Itˆ integral depends on the particular form of o the integral sums.
X(t) ∈ L2 (Ω.t4 ] E(F 2 (s))ds = 0. is a continuous process.12 X(t). Let t > s. T ]. Proposition 5. taking into account (5.11 Let F ∈ L2 ([0. Proposition 5. then that it is a continuous process. P. We have in fact.7) E[(X(t2 ) − X(t1 ))(X(t4 ) − X(t3 ))] t2 t4 The Itˆ integral o =E t1 T F (s)dB(s) t3 F (s)dB(s) T =E 0 T 1 [t1 . is mean square continuous. L2 (Ω)).68 Proof. We show now that X(t). l l We are going to show that X(t). T ] we have t E(X(t) − X(t0 )2 ) = t0 E(F (r)2 )dr . is a Ft –martingale Proof. T ]. T ]×Ω. Since t X(t) − X(s) = s F (r)dB(r). Ft . Then X ∈ CB ([0. Proof. dt×P). T ].t2 ] 1 [t3 . t0 ∈ [0. P).t4 ] F (s)dB(s) l = 0 1 [t1 . The conclusion follows. we have E[X(t)Fs ] = X(s) + E s t F (r)dB(r)Fs . t ∈ [0. . Moreover. For this we ﬁrst prove that it is a martingale with respect to the ﬁltration (Ft ) (see Appendix C). t ≥ 0. so that t→t0 lim E(X(t) − X(t0 )2 ) = 0. for any t. t ≥ 0. We know that for any t ∈ [0.t2 ] F (s)dB(s) l 0 1 [t3 .
(5. dt × P) t Fn (s)dB(s). P. Let (Fn ) ⊂ EB (0. because F (r) contains in general the “story” of the Brownian motion from 0 to r. t (5. We are now ready to prove the continuity of X.12) when F is an elementary process. T ) such that Fn → F and set Xn (t) = 0 (1) in L2 ([0. · · · . Theorem 5.T ] 0 EF (s)2 ds. F . dt × P) and let t X(t) = 0 F (s)dB(s). It is enough to prove (5. (5.12) is proved and the conclusion follows.ti ) . P. taking into account that Fs ⊂ Fi−1 . t ∈ [0. l where s = t1 . In this case.13 Let F ∈ L2 ([0. n ∈ N. . So. T ]. it remains to prove that t 69 E s F (r)dB(r)Fs = 0. t ∈ [0. n F = i=1 Fi−1 1 [ti−1 . T ] × Ω. tn = t and Fi−1 ∈ L2 (Ω.12) Notice that this is not obvious since s F (r)dB(r) is not independent of Fs in general (1) .13) 2 Proof. Then X has a continuous version and T E sup X(t)2 ≤ 4 t∈[0. T ] × Ω. we write t n E s n F (r)dB(r)Fs = i=1 E[Fi−1 (B(ti ) − B(ti−1 ))Fs ] = i=1 E{E[Fi−1 (B(ti ) − B(ti−1 ))Fi−1 ]Fs } = 0. T ].Chapter 5 So. since Fi−1 is Fi−1 –measurable and B(ti ) − B(ti−1 ) is independent of Fi−1 . P).
T ]) for almost all ω and its limit. Then W (τ ) is Fτ measurable and W (t + τ ) − W (τ ). P) is called a stopping time with respect to the ﬁltration (Ft )t≥0 if {τ ≤ t} ∈ Ft for all t ≥ 0. Consequently (Xn )(ω) is Cauchy in C([0. F. Taking into account Proposition 5. F.1 Itˆ integral with stopping times o Stopping times We proceed here as in Section 4.8 and 4. To any stopping time τ we associate the σalgebra Fτ : = {A ∈ F : A ∩ {τ ≤ t} ∈ Ft for all t ≥ 0}. Then by Corollary C. So.12 we see that X(t). P).8. is a continuous Ft –martingale. which coincides with X(ω) is continuous. T ]. they will be omitted. Then there exists a decreasing sequence (τn ) of discrete stopping times convergent pointwise to τ such that Fτn ⊃ Fτ for all n ∈ N. Proposition 5.4 5. The proofs of the two following propositions are completely similar to that of Proposition 4. A nonnegative extended random variable τ in (Ω.4. ω ∈ Ω.70 The Itˆ integral o Since B(t) is continuous it is clear that Xn (t) is continuous for all n ∈ N. t ≥ 0 is a Brownian motion in (Ω. 5. m ∈ N E sup Xn (t) − Xm (t)2 t∈[0.2.T ] T ≤ 4E(Xn (T ) − Xm (T )2 ) = 4E 0 Fn (s) − Fm (s)2 ds . .6 it follows that for any n. t ∈ [0.14 Let τ be a stopping time. Proposition 5.15 Let τ be a stopping time and set W (τ )(ω) = W (τ (ω))(ω).
0 where X(τ. P. t ∈ [0.16 Let F ∈ L2 ([0. Deﬁne τ F (s)dB(s) : = X(τ ). ω). P. n. ω) = X(τ (ω). T ] × Ω. has a continuous version. one can see that X(τ ) is Fτ –measurable. T ]. T ] × Ω. ¯ Consider now the stochastic process h(s) = 1 {s≤τ } .. s ∈ [0..14) Proof. .. If s ∈ [t1 .. Set Ai := {τ = ti }. l (5. Arguing as in Proposition 5.. .Chapter 5 71 5. dt × P) and let τ ≤ T be a stopping time. l We have h(s) = 1. T ]..2 Itˆ’s integral with stopping times o t Let F ∈ L2 ([0. Let moreover τ ≤ T be a stopping time. λ × P) and set X(t) = 0 F (s)dB(s).. Then Ai ∈ Fti . s ∈ [0. It is enough to prove the result when τ is of the form. T ]. t1 ). The following result reduces a Itˆ’s integral with a stopping time to a o usual one between 0 to T . tn ).. n. t ∈ [0. ω ∈ Ω. . Proposition 5. t2 ) we have h(s)(ω) = 1 if ω ∈ A2 ∪ · · · ∪ An . i = 1. with 0 < t1 < t2 < · · · < tn ≤ T .15 and using the fact that X(t). .. τ (Ω) = (t1 . t2 .4. Then we have τ T F (s)dB(s) = 0 0 1 {s<τ } F (s)dB(s). i = 1.
(5. l l 1 Similarly. P..15) Let now F ∈ L2 ([0. dt × P). Rd ) (that is such that any matrix element belongs to L2 ([0.5 Multidimensional Itˆ integrals o B(t) = (B1 (t).72 so that h(s) = 1 A2 ∪···∪An = 1 Ac .. L(Rm . dt×P. m.. j = 1.T ] be the natural ﬁltration of B (augmented with all Pnull sets of Ω) . d. l 5..j (t)dBj (t). Rd )).. L(Rm . l Then h is predictable and T t1 t2 The Itˆ integral o 1 {t<τ } F (s)dB(s) = l 0 0 F (s)dB(s) + 1 (A1 )c l t1 tn F (s)dB(s) + · · · + 1 (A1 ∪A2 ∪···∪An−1 l )c tn−1 F (s)dB(s) = X(t1 ) + 1 (A1 )c (X(t2 ) − X(t1 )) l + · · · + 1 (A1 ∪A2 ∪···∪An−1 )c (X(tn ) − X(tn−1 ) = X(τ ). . tk ) with k ≤ n we have h(s) = 1 (Ak ∪. if s ∈ [tk−1 . Then we have T T T E 0 f (s)dBi (s) 0 g(s)dBj (s) = δi. . dt × P. P). . g ∈ L2 ([0. Let (Ft )t∈[0. T ] × Ω. F . i = 1. We deﬁne the Itˆ o integral of F as the ddimensional process T m T F (t)dB(t) 0 i = j=1 0 Fi. First we need a lemma whose simple proof is left to the reader. i. T ] × Ω. . Rd )))... P.. T ]×Ω..∪An )c ...j 0 E[f (s)g(s)]ds. dt×P)). P. Lemma 5.17 Let f. We shall denote this space by L2 ([0. Bm (t)). t≥0 Let m ∈ N be ﬁxed and consider a standard mdimensional Brownian motion in the probability space (Ω. T ] × Ω. P. We shall deﬁne the Itˆ integral for predictable processes with values o in L(Rm .
(5. T ] × Ω. taking into account (5.16). Then we have m T (I(F ))i = j=1 0 Fi. Set I(F ) = T 0 F (t)dB(t). P. d m 0 T EI(F ) = i=1 j=1 2 E[Fi. d. dB(s) 0 and formula (5. Rd )). Remark 5. which yields (5. dB(t) = 0 EF (t)2 dt.16) where Tr denotes the trace. Proof.j (t)dBj (t). In this case we shall write the Itˆ integral of F as o T F (s). Then we have T 2 T E 0 F (t)dB(t) = 0 E[Tr (F (t)F ∗ (t))]dt. (5.18 Let F ∈ L2 ([0. · · · . i = 1.19 Assume that d = 1 so that L(Rd .15)... It follows that d m T 2 EI(F )2 = i=1 E j=1 0 Fi. Fm ). .j (t)dBj (t) and..Chapter 5 73 Proposition 5.17) . dt × P.16) reduces to T 2 T E 0 F (t). Rm ) is isomorphic to Rm and F becomes a vector F = (F1 . L(Rm .j (t)2 ]dt.
74 The Itˆ integral o .
For any k ∈ N we denote by Cb (R) the linear space of all real mappings which are uniformly continuous and bounded tok gether with their derivatives of order less or equal to k. t ≥ 0.Chapter 6 The Itˆ formula o 6. x∈R and k ϕ k = ϕ 0 + j=1 sup Dj ϕ(x). F . dt × P) and consider the stochastic process t t X(t) = x + 0 b(s)ds + 0 σ(s)dB(s). P) be a probability space. P. o Given a regular real function ϕ. σ ∈ L2 ([0. T ] × Ω.1 Introduction Let (Ω. We set dX(t) = b(t)dt + σ(t)dB(t) and call dX(t) the Itˆ diﬀerential of X. o k We need some notations. X is adapted. x∈R 75 . B a real Brownian motion. We are given two stochastic processes b. (Ft )t≥0 the natural ﬁltration of B augmented with the null sets of P and P the σalgebra of all predictable events (also augmented with the null sets of P). If ϕ ∈ Cb (R) we set ϕ 0 = sup ϕ(x).1) where x ∈ R. we are going to give a meaning to the Itˆ’s diﬀerential ϕ (X(t)). (6. continuous and continuous in mean square.
also as ϕ (X(t)) = ϕ (X(t))dX(t) + 1 2 σ (t)ϕ (X(t))dt.5) Proof. Write dX = b(t)dt + σ(t)dB and dϕ(X) = ϕ(X + dX) − ϕ(X) = ϕ (X)dX + = ϕ (X)dX + 1 2 1 2 ϕ (X)(dX)2 ϕ (X)b2 (t)(dt)2 + 2b(t)σ(t)dt dB + σ 2 (t)(dB)2 .19. We shall write (6. T ). setting ϕ (X(t)) = ϕ (X(t))σ(t)dB(t). Lemma 6. Remark 6. + or. 2 (6. P) (6. L2 (Ω.2 below. 2 t ≥ 0. k=1 . Writing (dB)2 = dt is justiﬁed by Lemma 6. Tthe following result on quadratic sums of a process is a generalization of Theorem 3. Put (dB)2 = dt and neglet the terms of order greater than dt.2) 1 2 σ (s)ϕ (X(s)) + b(s)ϕ (X(s)) ds. F .76 We shall prove the following Itˆ’s formula. F .3) t ≥ 0. T ]. Then we have n η→0 T lim F (tk−1 )(B(tk ) − B(tk−1 )) = k=1 0 2 F (s)ds in L2 (Ω. that is terms with (dt)2 and dt dB(t). P)) and let η = {0 = t0 < t1 < · · · < tn = T } ∈ Σ(0.4) 1 2 σ (t)ϕ (X(t)) + b(t)ϕ (X(t)) dt. o t Chapter 6 ϕ(X(t)) = ϕ(x) + 0 t ϕ (X(s))σ(s)dB(s) (6. Set Jη := n F (tk−1 )(B(tk ) − B(tk−1 ))2 .1 One can deduce formally Itˆ’s formula by proceeding as folo lows. (6.2) in the diﬀerential form.2 Let F ∈ CB ([0. 2 + 0 t ≥ 0.
6) write E Jη − n 2 2 F (tk−1 )(tk − tk−1 ) k=1 = E n F (tk−1 ) B(tk ) − B(tk−1 )2 − (tk − tk−1 ) k=1 n = k=1 E F (tk−1 )2 B(tk ) − B(tk−1 )2 − (tk − tk−1 ) n 2 +2 j<k=1 E F (tj−1 )[B(tj ) − B(tj−1 )2 − (tj − tj−1 )] F (tk−1 )[B(tk ) − B(tk−1 )2 − (tk − tk−1 )] Since the Brownian motion has independent increments.6) lim E Jη − k=1 F (tk−1 )(tk − tk−1 ) since. so that n 2 E Jη − k=1 n F (tk−1 )(tk − tk−1 ) = k=1 n E F (tk−1 )2 B(tk ) − B(tk−1 )2 − (tk − tk−1 ) 2 (6. . F . the last sum vanishes. obviously n η→0 T lim F (tk−1 )(tk − tk−1 ) = k=1 0 F (s)ds in L2 (Ω. To prove (6. (6.The Itˆ formula o It is enough to prove that η→0 77 n 2 = 0.7) = k=1 EF (tk−1 )2 E B(tk ) − B(tk−1 )2 − (tk − tk−1 ) 2 . P).
b and σ given by (6.78 since F (tk−1 ) and B(tk ) − B(tk−1 ) are independent. σ(t) = σ0 . t] with t ≤ λ1 . Then we obviously have N ϕ(X(t)) − ϕ(x) = k=1 [ϕ(X(tk )) − ϕ(X(tk−1 ))]. p − 1. .2) in [0.. Let η = {t0 = 0 < t1 < · · · < tN = t}..λi ) . 1.λi ) . 2 Lemma 6. We start by proving (6. . 3 2 Proof. 0 = λ0 < λ1 < · · · < λp and bi .. t ∈ [0. t ∈ [0.2) when 3 ϕ ∈ Cb (R). Now. In this case we have b(t) = b0 .1).3 Let ϕ ∈ Cb (R). taking into account that E[B(tk ) − B(tk−1 )2 ] = (tk − tk−1 ).8) and X by (6. E[B(tk ) − B(tk−1 )4 ] = 3(tk − tk−1 )2 . p p b= i=1 bi−1 1 [λi−1 . The conclusion follows.2) holds. l (6. Since Cb (R) is dense in Cb (R) it is enough to show (6. λ1 ] and X(t) = b0 t + σ0 B(t). σi are Fti measurable for any i = 0. as η → 0. l σ= i=1 σi−1 1 [λi−1 .8) where p ∈ N. First we assume that b o and σ are elementary processes. Now we are in position to prove Itˆ’s formula. x ∈ R. we have E Jη − k=1 n n 2 Chapter 6 F (tk−1 )(tk − tk−1 ) =2 k=1 E[F (tk−1 )2 ](tk − tk−1 )2 n ≤ 2η k=1 E[F (tk−1 )2 (tk − tk−1 )] → 0. λ1 ]. Then identity (6.
using Taylor’s formula we can write N 79 ϕ(X(t)) − ϕ(x) = 1 2 k=1 N ϕ (X(tk−1 ))(X(tk ) − X(tk−1 )) ϕ (X(tk−1 ))(X(tk ) − X(tk−1 ))2 + Rη k=1 + =: I1 + I2 + I3 .11) It is easy to check that lim I2.2 = 0 in L1 (Ω. P).The Itˆ formula o On the other hand.9) Concerning I1 we have N I1 = k=1 ϕ (X(tk−1 ))(b0 (tk − tk−1 ) + σ0 (B(tk ) − B(tk−1 )).1 + I2. P) η→0 η→0 (6. (6. (6.12) In fact I2.3 . So. F .1  ≤ 1 ϕ 2 b0 2 (tk − tk−1 )2 → 0 as η → 0 2 k=1 N .2 + I2.10) Concerning I2 we write N 2I2 = k=1 ϕ (X(tk−1 ))b2 (tk − tk−1 )2 0 N +2 k=1 N ϕ (X(tk−1 ))b0 σ0 (tk −k−1 )(B(tk ) − B(tk−1 )) + k=1 2 ϕ (X(tk−1 ))σ0 (B(tk ) − B(tk−1 ))2 =: I2. F . t η→0 t lim I1 = 0 ϕ (X(s))b(s)ds + 0 ϕ (X(s))σ(s)dB(s) in L2 (Ω. (6.1 = lim I2.
We have N 1 Rη = k=1 0 (1 − ξ)[ϕ (ξk ) − ϕ (X(tk−1 ))](X(tk ) − X(tk−1 ))2 dξ. (6.2 it follows that t η→0 lim 2I2.13) So. F . the conclusion will follow provided η→0 lim ERη  = 0. so that.80 and (1) N Chapter 6 EI2. we deduce setting 1 − ξ ≤ 1.14). (6. Moreover.2  ≤ ϕ 2 b0  σ0  k=1 N (tk − tk−1 )EB(tk ) − B(tk−1 ) ≤ ϕ 2 b0  σ0  k=1 (tk − tk−1 )3/2 → 0 as η → 0. where ξk = (1 − ξ)X(tk−1 ) + ξX(tk ). P). 3 Since ϕ ∈ Cb (R) we have by the mean value theorem. . by Lemma 6.14) Let us prove (6. ϕ (ξk ) − ϕ (X(tk−1 )) ≤ ϕ 0 (1 − ξ)X(tk ) − X(tk−1 ). N Rη  ≤ ϕ Consequently N 3 k=1 X(tk ) − X(tk−1 )3 .3 = 0 ϕ (X(s))σ 2 (s)ds in L2 (Ω. N Rη  ≤ 3 ϕ 3 b0 3 k=1 (1) tk − tk−1 3 + 3 ϕ 3 σ0 3 k=1 B(tk ) − B(tk−1 )3 since EB(t) ≤ [EB 2 (t)]1/2 = t1/2 .
. N 3 k=1 E(Rη ) ≤ 3 ϕ 3 b0  tk − tk−1  + 3 ϕ 3 σ0  3 3 √ N 15 k=1 tk − tk−1 3/2 → 0. We ﬁnally prove 2 Theorem 6. for any j ∈ N. The proof is complete when t ≤ λ1 . Set. j→∞ lim σj = σ in L2 ([0. P.10)) j→∞ lim Xj = X in CB ([0. (6. t t Xj (t) = x + 0 bj (s)ds + 0 σj (s)dB(s). as η → 0. σ ∈ L2 ([0. dt × P) and ϕ ∈ Cb (R). The general case can be treated in the same way taking into account that bk−1 and σk−1 are independent of B(tk ) − B(tk−1 ). L2 (Ω)).The Itˆ formula o and so (2) 81 . Then identity (6. Proof. T ].2) holds for all t ∈ [0.4 Let x ∈ R. Taking expectation in the Itˆ formula we ﬁnd a useful identity which o allows to estimate the expectation of ϕ(X(t)). (6. P. Let (bj ) and (σj ) be sequences of elementary processes such that j→∞ lim bj = b. T ] × Ω.16) 1 σj (s)ϕ (Xj (s)) + bj (s)ϕ (Xj (s)) ds. s ∈ [0. T ] × Ω. dt × P). b.15) Then we have (see (5. (2) Since EB(t)3 ) ≤ [E(B(t)6 )]1/2 = √ 15. T ]. Moreover by (6.2) we have t ϕ(Xj (t)) = ϕ(x) + 0 t ϕ (Xj (s))σj (s)dB(s). 2 + 0 Now the conclusion follows by the dominated convergence theorem letting j → ∞. T ].
t ∈ [0.17) holds. Set t t X(t) = x + 0 b(s)ds + 0 σ(s)dB(s). Example 6. Then condition (6. σ ∈ L2 ([0. P.1 The Itˆ formula for unbounded functions o We want now to show that formula (6. (6.18) and assume in addition that t E 0 ϕ (X(s))σ 2 (s) + 2ϕ (X(s))b(s)ds < +∞.82 Chapter 6 Proposition 6.6 Assume that x ∈ R. T ] × Ω. Let t t X(t) = x + 0 b(s)ds + 0 σ(s)dB(s). ϕR (x) = 0 if x ≥ R + 1. b.5 Assume that x ∈ R. Proposition 6.19) Then E[ϕ(X(t))] < +∞ and (6. P. Then E[ϕ(X(t))] = ϕ(x) + 1 E 2 t [ϕ (X(s))σ 2 (s) + 2ϕ (X(s))b(s)]ds. T ] × Ω.17) 6.17) also holds without the assumption that ϕ is bounded. provided the integrand in the right hand side is summable. Then t E(X(t)2 ) = x2 + E 0 (σ 2 (s) + 2X(s)b(s))ds. 0 (6.6.1.19) becomes t E 0 σ 2 (s) + 2X(s)b(s)ds < +∞ which is clearly fulﬁlled. . T ]. σ ∈ L2 ([0. dt × P) and 2 ϕ ∈ Cb (R). 2 Cb (R) Proof of Proposition 6. b.7 Take ϕ(x) = x2 . For any R > 0 consider a function ϕR ∈ such that ϕ(x) if x ≤ R. (6. T ]. t ∈ [0. dt × P) and ϕ ∈ C 2 (R).
l 2 0 Now.19).T ] T if sup X(t) < R.s.. m > 1. by the assumption (6.23) 1 s<(t∧τR ) [ϕ (X(s))σ (s) + 2ϕ (X(s)b(s)]ds.The Itˆ formula o Then. As an application of Proposition 6. applying Itˆ’s formula (6. Let now τR be the stopping time inf{t ∈ [0. l (6. T 0 2m F (s)dB(s) .T ] τR = It is clear that τR is increasing and bounded by T .22) Taking expectation we obtain E[ϕ(X(t ∧ τR ))] − ϕ(x) 1 = E 2 t (6. R→∞ lim τR = T P–a.16 we can write ϕ(X(t ∧ τR )) − ϕ(x) = 1 2 + 0 t 1 s<(t∧τR ) [ϕ (X(s))σ 2 (s) + 2ϕ (X(s)b(s)]ds l 0 t 1 s<(t∧τR ) ϕ (X(s)))σ(s)dB(s). t∈[0. We know that X(·. t∈[0.6 let us estimate E where F is predictable and m ∈ N. say M (ω). in view of Proposition 5. we can let R → ∞ obtaining the conclusion. For such a ω. X(·. yields for any R > 0 o ϕR (X(t)) − ϕ(x) = 1 2 + 0 t 83 [ϕR (X(s))σ 2 (s) + 2ϕR (X(s)b(s)]ds 0 (6.21) Now. (6. T ] : X(t) ≥ R} if sup X(t) ≥ R.21) and the dominated convergence theorem. ω) attains the maximum. So.2) to ϕR (X(t)). (6. ω) is continuous for almost all ω ∈ Ω. Then we have τR (ω) = T for all R > M (ω).20) t ϕR (X(s)))σ(s)dB(s).
Set t t X(t) = x + 0 b(s)ds + 0 σ(s)dW (s). dt × P)). Substituting this in (6. dt × P) and we have T E[X(T ) 2m ] ≤ [m(2m − 1)] T m m−1 0 E F (t)2m dt. (6. T ]. Rd )).19) holds so that. 4 and so on. and set t X(t) = 0 F (s)dB(s).2 Itˆ’ formula for a vector valued process o Let d. (6. P. b ∈ L2 ([0. dt × P. m ∈ N. (6. T ] × Ω. dt × P).26) From which 0 EX(t)4 dt ≤ 36T 2 0 EF (t)4 dt. So. by Proposition 6. P. We start from the case m = 2. P.84 Chapter 6 Proposition 6.6 we have t E[X(t)4 ] = 6E 0 X(s)2 F (s)2 ds . m ∈ N. T ] × Ω. T ] × Ω.24) Proof. Assume that x ∈ Rd . dt × P) is dense in L2m ([0. setting ϕ(x) = x4 . P. Then X ∈ L2m ([0.8 Assume that F ∈ L2m ([0.25) yields T E[X(t)4 ] ≤ 36T E 0 F (t)4 dt. We can now easily iterate the previous argument taking successively m = 3. L(Rm . yields T T 1/2 T 1/2 EX(t) dt ≤ 6T E 0 0 T 4 X(t) dt T 4 E 0 F (t) dt 4 . T ] × Ω. It is enough to prove (6. dt × P.25) Integrating between 0 and T .24) is proved for m = 2. P. T ] × Ω. t ∈ [0. By H¨lder’s inequality it follows that o t 1/2 t 1/2 E[X(t)4 ] ≤ 6 E 0 X(s)4 ds E 0 F (s)4 ds . T ] × Ω. (6. t ∈ [0. Then (6. 6. P. Rd ) and σ ∈ L2 ([0.24) when F is bounded (because L∞ ([0. T ] .
27) in the diﬀerential form ϕ (X(t)) = Dϕ(X(t)). Then we have n σ→0 lim f (tk−1 )(Bi (tk ) − Bi (tk−1 ))(Bj (tk ) − Bj (tk−1 )) k=1 T (6. Then we have n σ E[(Ii.j 0 Proof. σ(t)dB(t) + 1 Tr[(σσ ∗ )(t)D2 ϕ(X(t))] + b(t). Let i = j and set n η Ii... Dϕ(X(t)) 2 dt. (6.j := k=1 f (tk−1 )(Bi (tk ) − Bi (tk−1 ))(Bj (tk ) − Bj (tk−1 )). Let η = {0 = t0 < t1 < · · · < tn = T } be a decomposition of [0. in L2 (Ω.j )2 ] = E h.k=1 f (th−1 )f (tk−1 )(Bi (th ) − Bi (th−1 ))(Bj (th ) − Bj (th−1 )) × (Bi (tk ) − Bi (tk−1 ))(Bj (tk ) − Bj (tk−1 )) n =E h=1 f 2 (th−1 )(Bi (th ) − Bi (th−1 ))2 (Bj (th ) − Bj (th−1 ))2 n = h=1 E(f 2 (th−1 ))(th − th−1 )2 → 0.9 Let f ∈ CB ([0.27) for all t ∈ [0. (6.28) The proof is similar to that of the onedimensional case seen before. 2. Dϕ(X(s)) 2 + 0 ds. j ∈ {1. We shall write (6. 1 Tr[(σσ ∗ )(s)D2 ϕ(X(s))] + b(s). o t 85 ϕ(X(t)) = ϕ(x) + 0 t Dϕ(X(s)). So. L2 (Ω)) and let i. = δi.29) follows from Lemma 6..The Itˆ formula o We are going to prove the following Itˆ’s formula. Let us start with a preliminary lemma. we shall only sketch some points of the proof. T ].2. T ]. P).29) f (s)ds. (6. t ≥ 0. F . m}. σ(s)dB(s) . If i = j. . Lemma 6. T ].
λ1 ] t ∈ [0. Let η = {t0 = 0 < t1 < · · · < tN = t}. Fti .30). P. l (6. (6. k for all x. We have b(t) = b0 . 3 Proof. X(t) = b0 t + σ0 B(t). X(tk ) − X(tk−1 ) + 1 2 N D2 ϕ(X(tk−1 ))(X(tk ) − X(tk−1 )). (3) On the other hand. b0 (tk − tk−1 ) + σ0 (B(tk ) − B(tk−1 ) . .31) Concerning I1 we have N I1 = k=1 (3) Dϕ(X(tk−1 )).λi ) . o p p b= i=1 bi−1 1 [λi−1 .10 Let ϕ ∈ Cb (Rd ).3. h. λ1 ].86 Chapter 6 as σ → 0. Rd )) i = 0. We use the notations Dϕ(x)h = Dϕ(x). Then we obviously have N ϕ(X(t)) − ϕ(x) = k=1 [ϕ(X(tk )) − ϕ(X(tk−1 ))]. P. Then identity (6. Fti . taking ϕ ∈ Cb (Rd ) and proving (6. by Taylor’s formula we can write N ϕ(X(t)) − ϕ(x) = k=1 Dϕ(X(tk−1 )). Rd ) and σi ∈ L2 (Ω. x ∈ Rd and let b and σ given by (6.6) in [0.. t] with t ≤ λ1 . 0 = λ0 < λ1 < · · · < λp bi ∈ L2 (Ω. 1. Now we prove Itˆ’s formula when b and σ are elementary processes as. We proceed as in the proof of Lemma 6. p − 1. L(Rm . h and D2 ϕ(x)(h. 2 Lemma 6.. k) = D2 ϕ(x)h. . l σ= i=1 σi−1 1 [λi−1 .30) where p ∈ N. X(tk ) − X(tk−1 ) + Rη k=1 =: I1 + I2 + I3 . and σ(t) = σ0 .. t ∈ [0.27) holds.λi ) . k ∈ Rd .
P).32) Concerning I2 we write N 2I2 = k=1 D2 ϕ(X(tk−1 ))b0 .3 = = D2 ϕ(X(tk−1 ))(σ(B(tk ) − B(tk−1 ))). t η→0 t 87 lim I1 = 0 Dϕ(X(s)). σ0 (B(tk )−B(tk−1 )) =: I2.β=1 Therefore.2 = 0 in L1 (Ω.The Itˆ formula o So.2 +I2.1 +I2.β (s)ds 0 i. proceeding as before. σ(s)dB(s) in L2 (Ω. taking into account Lemma 6. (6. b0 (tk − tk−1 )2 N +2 k=1 N D2 ϕ(X(tk−1 ))b0 . Now. P) η→0 (6. (6.j ϕ σi. σ0 (B(tk ) − B(tk−1 )) (tk − tk−1 ) + k=1 D2 ϕ(X(tk−1 ))σ0 (B(tk )−B(tk−1 )).j ϕ(X(s)) σi. we have N 2I2.1 = lim I2.3 = = 0 Tr [D2 ϕ(X(s))(σσ ∗ (s))]ds. (6.34) Moreover. F .α (Bα (tk ) − Bα (tk−1 )) σi.α (s)σi. F . σ(B(tk ) − B(tk−1 )) k=1 N d m 2 Di.35) .j=1 α=1 t lim 2I2. b(s) ds+ 0 Dϕ(X(s)).j=1 α.33) It is easy to check that η→0 lim I2.9 we have t η→0 d m 2 Di. we see that η→0 lim ERη  = 0.β (Bβ (tk ) − Bβ (tk−1 )).3 . k=1 i.
Prove that 1 ϕ (X(t))σ(t)2 dt. 2... 2 = . 2 (6. dt × P)...13 Let d ∈ N. P. σi ∈ L2 ([0. P. T ] × Ω. Rd )). Exercise 6. i = 1. The general case can be treated in the same way taking into account that bk−1 and σk−1 are independent of B(tk ) − B(tk−1 ). P. dt × P : Rd ). σk ∈ L2 ([0. k = 1. Let ϕ ∈ 2 Cb (R). Finally. σm (t)).. m ∈ N. T ]. . dX(t) + 1 D2 ϕ(X(t))σ(t). b. Let moreover ϕ ∈ Cb (Rd ).. 2 where σ = (σ1 . . Prove that dϕ(X(t)) = Dϕ(X(t)). proceeding as we did for the proof of Theorem 6. T ] × Ω. σ(t) dt.. m = 1 bi .27) holds for any t ∈ [0...11 Let b ∈ L2 ([0. Then identity (6. Exercise 6. i = 1. T ] × 2 Ω. x ∈ Rd and ϕ ∈ Cb (Rd ). σd ). m. dt × P)..4 we obtain the result Theorem 6. σ ∈ L2 ([0.12 Let d = 1. Set m t t X(t) = 0 b(s)ds + k=1 0 σk (s)dBk (s).36) dϕ(X(t)) = ϕ (X(t))dX(t) + where σ(t) = (σ1 (t).. d..88 Chapter 6 The proof is complete when t ≤ λ1 . Set X(t) = b(t)dt + σdB(t).37) . dt × P : L(Rm . . 2 (6. T ] × Ω. P.
L(R . b is called the drift and σ the diﬀusion coeﬃcient of the equation. We denote by (Ft )t≥0 the natural ﬁltration of B(t) (augmented with all Pnull sets of Ω). L2 (Ω. T ].1).3) 89 . (7. T ). (7. L (Ω.Chapter 7 Stochastic evolution equations We are given two positive integers r. Let us consider the following integral equation t t X(t) = η + s b(u. t ≥ 0. R ))) and 0 ≤ a < b ≤ T. Rd ) G(t) 2 HS E a G(t)dB(t) = a E dt. P. By a solution of equation (7. (7. This suggests to endow L(Rr . We shall write (7.1) where s ∈ [0. d and an rdimensional standard Brownian motion B(t). η ∈ L2 (Ω. F . X(u))du + s σ(u. for all G ∈ CB ([0. X(t))dt + σ(t.2) X(s) = η. Rd ) with the Hilbert–Schmidt norm. t ∈ [s. Rd ). T ]. based on the identity b 2 b E a G(t)dB(t) 2 r = a d E [Tr (G(t)G∗ (t))] dt. in a probability space (Ω. b : [0. T ]. Fs . setting S and to write b HS : = [Tr(SS ∗ )]1/2 . X(t))dB(t). Rd ).1) on the interval [s. T ] × Rd → L(Rr . In order to solve (7. T ] we mean a function X ∈ CB ([s. P).1) we shall use a ﬁxed point argument. 2 b S ∈ L(Rr . X(u))dB(u). T ] × Rd → Rd and σ : [0.1) in diﬀerential form as dX(t) = b(t. Rd )) that fulﬁlls equation (7.
X(u))dB(u). T ].1 holds and let s ∈ [0. x) 2 HS 2 HS ≤ M 2 x − y2 (7. (7.4). Rd )). Rd ). Proof. after possibly changing the constant M .1) is equivalent to the following. γ1 and γ2 map CB into itself. s X ∈ CB . y) and b(t. L2 (Ω. We are going to solve (7. T ]. x.90 Chapter 7 7. σ(u. Step 1.5) is a consequence of (7. Rd )). T ]. x) − b(t.1) are the following. T ]. T ] γ2 (X)(t) := and set γ(X) := η + γ1 (X) + γ2 (X).5) Notice that. η ∈ L2 (Ω. Hypothesis 7. Fs .4) ≤ M 2 (1 + x2 ). X(u))du.1) by a ﬁxed point argument in the space CB := CB ([s. (7. t ∈ [s. x)2 + σ(t. Then problem (7. t ∈ [s. we have b(t. T ).6) . (ii) There exists M > 0 such that for all t ∈ [0.1 (i) b and σ are continuous on [0. P. y)2 + σ(t. Then equation (7. L2 (Ω. (7. y ∈ Rd .1 Assume that Hypothesis 7. Deﬁne γ1 (X)(t) := s t t b(u. T ] × Rd .1) has a unique solution X ∈ CB ([s. Theorem 7. x) − σ(t. X = η + γ1 (X) + γ2 (X) = γ(X). X ∈ CB . X ∈ CB .1 Existence and uniqueness The standard assumptions for the wellposedness of problem (7.
we see that γ2 maps CB into itself. Let X. using the H¨lder inequality and taking into aco count (7. γ1 (X) − γ1 (Y ) Furthermore CB ≤ M (T − s) X − Y CB . X(u)) 2 HS )du ≤ M2 s (1 + X(u)2 )du ≤ M 2 (t − s)(1 + X 2 CB ) So. Y (u)) 2 HS )du 2 CB . t γ1 (X)(t) − γ1 (Y )(t)2 ≤ (t − s) s t 2 s b(u. Concerning γ2 we have taking into account (7. We have. Y ∈ CB . ≤ M 2 (t − s) X − Y . (1 + X(u)2 )du ≤ M 2 (t − s)2 (1 + X Since γ1 (X)(t) is Ft –measurable for all t ∈ [s. using again the H¨lder inequality and taking o into account (7. t t γ1 (X)(t)2 ≤ (t − s) s b(u. T ]. t Eγ2 (X)(t) = s t 2 E( σ(u.3) and (7. X(u))2 du ≤ M 2 (t − s) s 2 CB ). X(u)) − b(u. γ1 maps CB into itself and γ1 (X) CB ≤ M (t − s)(1 + X CB ). X(u)) − σ(u. Y (u))2 du ≤ (t − s)M Consequently X(u) − Y (u)2 du ≤ (t − s)2 M 2 X − Y 2 CB du.5).7) t Eγ2 (X)(t) − γ2 (Y )(t)2 = s E( σ(u.Stochastic evolution equations 91 Concerning γ1 we have. γ is Lipschitz continuous.4). X. Step 2.5). Y ∈ CB (7.
s. Whe shall use greek letters for stochastic initial data and latin letters for deterministic ones. t ∈ [s. .1 it follows that Z(t) = X(t.3 Assume that Hypothesis 7. (7. it possesses a unique ﬁxed point. Proposition 7.10) Proof. By the uniqueness part of Theorem 7. T ]. Rd ). Then Z solves the problem dZ(t) = b(t. s.1) on [s. Z(t))dt + σ(t. η). η) = X(t. as required. Let us prove the cocycle law. X(r. Then by the previous argument there is a unique solution to (7. r. Y ∈ CB . s. s. and so. Now we repeat the proof with T1 replacing s and in a ﬁnite number of steps we arrive to the conclusion. Y ∈ CB . η)). X(r.92 and so. r. X. η) which belongs to L2 (Ω. Now if T − s is such that √ M T − s + T − s ≤ 1/2. If (7.9) γ is a 1/2–contraction on CB .8) By (7. 0 ≤ s ≤ r ≤ t ≤ T. η)).1). s. CB . s. s. Z(t))dB(t).2 By Theorem 5. In the following we shall denote by X(·. Fs . Remark 7.8) it follows that γ maps CB into itself and √ γ(X) − γ(Y ) CB ≤ M (T − s + T − s ) X − Y  for all X. C([s.9) does not hold we choose T1 ∈ (s. γ2 (X) − γ2 (Y ) CB Chapter 7 ≤M √ T −s X −Y CB . Then X(t.1 holds and let η ∈ L2 (Ω. Z(r) = X(r. T ] such that M T1 − s + T1 − s ≤ 1/2. s. η) the solution of problem (7. η).7) and (7. T ])) and so it is a continuous process. Deﬁne Z(t) = X(t. η) = X(t. P. (7. (7. T1 ].13 it follows that there exists a version of the solution X(·.
s. T ].. Then we have b(u. η) = X(·. s.14) Proof. More precisely. We claim that n XN (t. Fs . .1) can be obtained as a limit of successive approximations. s.. and A1 . Let XN be deﬁned by (7.15) is clear for N = 0.11). s.1 holds and that n η= k=1 xk 1 Ak . η) = η + s b(u. η) = k=1 XN (t.. gives some information about the relationship between X(t.. s. η) in CB ([s. deﬁne X0 (t. σ(u. Rd )). . η) = η and for any N ∈ N. s. XN (u. xn ∈ Rd . x). (7. xk ) in Ak . η)) = b(u. XN (u. An are mutually disjoints sets in Fs such that n Ω= k=1 Ak .15) Once (7. XN (u... s. η) = XN (t. P.Stochastic evolution equations 93 Remark 7. s. . s. k = 1.4 By the contraction principle it follows that the solution X(t. XN (u. xk )1 Ak .. Rd ) and X(t. L2 (Ω. . η) = n X(t.. . XN (u. xk )1 Ak . l k=1 (7. n.. (7. s. s.. l ∀ N ∈ N. η))dB(u). Then we have X(t. s. η) of problem (7. η)) = σ(u..5 Assume that Hypothesis 7. so that XN (t.13) where x1 . xk )) in Ak . s. Proposition 7. η))du + s σ(u. k = 1. n. Equality (7. XN (u.11) Then we have N →∞ lim XN (·.. s.12) Next result. . η). t t XN +1 (t. which as we shall see plays an important rˆle in proving that o X(·. s. s.15) is proved. k = 1... s. l (7. s. x) is a Markov process. Let us proceed by recurrence. xk )) in Ak . s. n. (7. the conclusion follows letting N tend to inﬁnity. s. x ∈ Rd . Assume that it holds for a given N ∈ N. η ∈ L2 (Ω..
η)) = k=1 1 Ak σ(u. Rd ) and x ∈ Rd .1. s. XN (u. L2m (Ω. T ]. s. 7. x) ∈ CB ([s. s. s.24) proved in Proposition 6.6 Assume that Hypothesis 7. xk ))dB(u) = k=1 1 Ak XN +1 (t. s. using inequality (6. T ]. Fs .1 holds and let m ∈ N. s. L2m (Ω. T ]. XN (u. l σ(u. η) ∈ CB ([s. xk ) l and (7. (7.7 Consider the stochastic diﬀerential equation dX = AXdt + CdB(t). Rd ). XN (u. L2m (Ω. C ∈ L(Rr . s.8. xk )). xk )). Rd )). where A ∈ L(Rd ). s.94 so that b(u.1 Solution of the stochastic diﬀerential equation in the space CB ([s. η)) = k=1 n Chapter 7 n 1 Ak b(u. s. Rd )). s. s. Rd )). T ]. XN (u. s ∈ [0. 7. XN (u.15) holds for N + 1. P.2 Examples Example 7. η) = k=1 1 Ak X0 (t.1 by a ﬁxed point argument in the space m CB := CB ([s. X(0) = x. XN (u. the conclusion follows.1.1) has a unique solution X(·. In particular X(·. T ).16) . xk )du s n + s σ(u. l Consequently n t XN +1 (t. Proof. We proceed as in the proof of Theorem 7. Theorem 7. L2m (Ω. So. xk ) + l t b(u. η ∈ L2m (Ω. Rd )). ∀ x ∈ Rd . Then problem (7.
(7. we ﬁnd X(t) = e x + 0 tA e(t−s)A CdB(s).16) has a unique solution X(t) which fulﬁlls the integral equation t X(t) = x + A 0 X(s)ds + CB(t). We want to show that the solution of (7. T ]. We obtain t Y (t) = 0 e(t−s)A (x + CB(s))ds.8 Let r = d = 1 and consider the stochastic diﬀerential equation dX = aXdt + cXdB(t). thanks to Proposition 3. T ]. c. Again Theorem 7.19) is given by 1 2 X(t) = et(a− 2 c ) ecB(t) x.1 applies. (7. which can be easily solved by the method of variation of constants. x ∈ R.19).20) solves (7. Y (0) = 0. t ≥ 0. Write X(t) = eF (t) where F (t) = t a − 1 c2 + cB(t). X(0) = x.17) Setting t Y (t) = 0 X(s)ds. Taking into account that. (7. t ∈ [0. Y fulﬁlls the equation Y (t) = AY (t) + x + CB(t).19) where a. (7. T ].1 applies so that (7.18) Example 7.12.17) yields t X(t) = A 0 e(t−s)A (x + CB(s))ds + x + CB(t). t t e(t−s)A CdB(s) = CB(t) + A 0 0 t e(t−s)A CB(s)ds.Stochastic evolution equations 95 Clearly Theorem 7.20) For this we check that X(t) given by (7. Then we have 2 dF (t) = a− 1 2 c dt + cdB(t) 2 . t ∈ [0. t ∈ [0. By substituting Y (t) in (7.
4) one deals with stochastic diﬀerential equations having random coeﬃcients. t t X(t. (7. ω)dB(u). ω) = η(ω) + s b(u. L2 (Ω. o dX(t) = eF (t) dF (t) + = eF (t) Chapter 7 1 2 F (t) c e dt 2 1 1 a − c2 dt + cdB(t) + c2 eF (t) dt 2 2 = aX(t)dt + cX(t)dB(t). x. T ].24) ≤ M 2 (1 + x2 ). ω). for all t ∈ [0. ω) = σ(t. X(u. b : [0.1. Rd )) and V ∈ CB ([0. L(Rr . x. (7. T ]. ω) − σ(t. Exercise 7. ω ∈ Ω. ω) and b(t. . T ].96 and. The following result can be proved as Theorem 7. by Itˆ’s formula. Y (t. ω). x ∈ Rd and AC = CA. T ]. U (t. Rd ))) where. ω)du + s σ(u. ω)2 + σ(t. L2 (Ω. X(u. (7. T ] × L(Rr .21) where A.2 (i) There exists M > 0 such that for all t ∈ [0. ω) 2 HS 2 HS ≤ M 2 x − y2 (7. x. ω)2 + σ(t. ω). y ∈ Rd . C ∈ L(Rd ). ω)). Y (t. ω) = b(t. y. ω). Rd ). x.9 Let r = 1 and consider the diﬀerential stochastic equation dX = AXdt + CXdB(t). Fs . ω) − b(t. X(0) = x. Show that the solution of (7. Rd ) × Ω → Rd are such that: Hypothesis 7.1.25) (ii) For any Y ∈ CB ([0. (7. T ] × Rd × Ω → Rd and σ : [0. ω)).22) 7.23) Here η ∈ L2 (Ω. Rd )) we have U ∈ CB ([0.3 and 7. V (t. ω ∈ Ω b(t. y. T ]. x.3 Diﬀerential stochastic equations with random coeﬃcients In some situations (see Subsections 7.21) is given by 2 X(t) = et(A−C /2) eCB(t) x. L2 (Ω.
1) is H¨lder continuous on t. 2 Now by Itˆ’s formula we ﬁnd o dX(t) = eH(t) dH(t) + 1 2 1 2 t t F (s)2 ds + 0 0 F (s). Now it is easy to check that Theorem 7.dB(s) x. s and Lipschitz o continuous on η in mean square.Stochastic evolution equations 97 Theorem 7. Let us show that X(t) = e− 2 1 Rt 0 Rt F (s)2 ds+ 0 F (s). .10 Assume that Hypothesis 7. Example 7. dB(t) . Rd ). (7. First we show that EX(t. Rd )).27) is proved.2 holds. Then problem (7. s.1 holds.26). dB(t) = X(t) F (t). L2 (Ω. T ].2 7.1 Continuous dependence on data Continuous dependence on mean square We assume here that Hypothesis 7. dB(t) . = eH(t) F (t). 7.27) For this we check that X(t) given by (7. T ]. T ) and η ∈ L2 (Ω. Fs . We are going to prove that the solution X(t.26). Rd )). η) to (7. dB(t) . t ∈ [0.26) X(0) = x. T . η)2 is bounded. dB(s) . Let s ∈ [0. t ≥ 0.2. (7.27) solves (7. Write X(t) = eH(t) where H(t) = − Then we have 1 dH(t) = − F (t)2 dt + F (t). L∞ (Ω. where F ∈ CB (0. t ≥ 0. So. (7. eH(t) F (t)2 dt t ≥ 0.11 Let d = 1 and consider the stochastic diﬀerential equation dX(t) = X(t) F (t). s.10 applies and so there exists a solution X of (7.23) has a unique solution X ∈ CB ([s.
Proposition 7. E(η2 )) such that we have E X(t. We note that. 0 ≤ s < t ≤ T. X(u))du +3 s E( σ(u.12 Assume that Hypothesis 7. s. E(η2 ))(t − t1 ). Then there exists a constant C1 (T. s. P. s. η)2 ≤ C(T. Then for all s ∈ [0. by Lemma 7. Rd ).13 Assume that Hypothesis 7. η) − X(t1 . s. s. (7. Consequently E (X(t)2 ) ≤ 3E(η2 ) + 3M 2 ((T − s)2 + (T − s)) t +3M ((T − s) + 1) s 2 E X(u)2 du. there exists a constant C(T.1 holds. s. s. Writing for short X(t.98 Chapter 7 Lemma 7. η) with respect to t. Fs . we have t 2 2 (T −s+1) . E(η2 )) such that E X(t.1(ii) and the H¨lder inequality we deduce that o t E (X(t)2 ) ≤ 3E(η2 ) + 3M 2 (t − s) s t (1 + E X(u)2 )du +3M 2 s (1 + E X(u)2 )du. T ] and η ∈ L2 (Ω. η)2 ≤ 3[E(η2 ) + M 2 ((T − s)2 + (T − s)]e3M Proof.12.28) E (X(t) ) ≤ 3E(η ) + 3E s t 2 2 b(u. η)2 ≤ C1 (T. Rd ) we have E X(t. (7.1 holds. We now study the regularity of X(t. Fs . η) with respect to t.30) . By Hypothesis 7. Let 0 ≤ s ≤ t1 < t ≤ T and η ∈ L2 (Ω. E(η2 )). (7. The conclusion follows from the Gronwall lemma. s. η. η) = X(t). X(u)) 2 HS )du.29) We start with the regularity of X(t.
ζ)2 ≤ 3η − ζ2 t 2 2 (T −s+1)(t−s) E(η − ζ2 ). X(u. We ﬁnally study the regularity of X(t. s. (1 + E X(u. E X(t.4) we obtain E (X(t. s. s. X(u.η > 0 such that E X(t. let 0 < s < s1 < t ≤ T. Let us study the regularity of X(t. let 0 ≤ s < t ≤ T and η.1 holds. s.31) +3 s (b(u. ζ)2 du and the conclusion follows from the Gronwall lemma. ζ)2 ) ≤ 3E(η − ζ2 ) + 3M 2 (T − s + 1) t × s E X(u. Then E X(t. s. and η ∈ L2 (Ω. η)2 ≤ CT. η) − X(t.15 Assume that Hypothesis 7. (7. η) − X(t. η) − b(u. s1 . s. Rd ). Proposition 7. s. η)2 )du.Stochastic evolution equations Proof. We have X(t. s.14 Assume that Hypothesis 7. η)2 ≤ 2M 2 ((t − t1 )2 + t − t1 )(1 + C 2 (T. s.η s − s1 . s. η) − X(t. η) − X(t. Fs . Then there exists a constant CT.1 holds. ζ))du t 2 +3 s (σ(u. s. s. s. X(u. P. η) − X(u. η)2 du t 2 t1 + 2M Consequently. η) with respect to s. We have t 99 E X(t. E(η2 ))) and the conclusion follows. Proposition 7. η)2 ≤ 2M 2 (t − t1 ) t1 (1 + E X(u. ζ)2 ≤ 3e3M Proof. η) − σ(u. ζ))dB(u) . s. η) − X(t1 .32) . Fs . (7. s. Taking expectation and using (7. s. η) − X(t1 . s. Rd ). s. s. η) with respect to η. s. ζ ∈ L2 (Ω. s. X(u.
x) such that E X(t.3 Almost sure continuity and h¨lderianity o of trajectories In this section we show that X(·. whose deﬁnition is recalled in Appendix E below. η) − X(s. s.3 and the Sobolev embedding theorem E. y)2m ≤ C(T )x − y2m . s1 . 1/2). First we need a lemma. s. x) − X(t1 .18 Assume that Hypothesis 7. s.10). s. X(·. x)2m < +∞. Taking into account the cocycle law (7. η) − X(t. η). 7. η) = X(t.13 using (6. which can be proved as Proposition 7.30).100 Proof. η) − η2 ) 2 = CT E (X(s1 . Lemma 7. x ∈ Rd and m ∈ N.33) Now from Proposition E. (7.16 Assume that Hypothesis 7.34) ([s. Then we have E X(·. we can write Chapter 7 X(t. Let x ∈ Rd . X(s1 . ·). s. Then there is a constant C(T ) > 0 such that E X(t. η) − X(t. η)2 ) ≤ CT E (X(s1 . η)2 ) .1 holds. we consider almost sure regularity of X(t. x) belongs to C −1/(2m) (7. s1 . s. 0 ≤ s ≤ t ≤ T . The conclusion follows now from (7. s1 . y ∈ Rd .17 Assume that Hypothesis 7.35) . Let 0 ≤ s ≤ t1 < t ≤ T. s. s. x)2m ≤ C1 (T. Then there exists a constant C1 (T. s. arguing as in the proof of Proposition 7. s. s. x2 ))(t − t1 )m . By (7. s1 . First. let 0 ≤ s < t ≤ T and x. s.1 it follows that Proposition 7. Finally. s. s. x) − X(t. m ∈ N and ∈ (0. x) is H¨lder o continuous almost surely. s. T ]) almost surely.2m Moreover. .1 holds. (7. Then the Sobolev embedding theorem (also stated in Appendix E) will imply that X(·. x) belongs to a suitable Sobolev space. η)) − X(t.14 we have Lemma 7.24).31) there exists CT > 0 such that 2 E (X(t.1 holds.
4 Diﬀerentiability of X(t. . s. X(t. t∈[0.2m Moreover. s. x → X(·. Then for any m > 1 and ∈ (0. T ]) =: CB ([s. T ]. 1]d ) almost surely.3 it follows that 101 Proposition 7.1 Existence of Xx (t.3 2 2 (i) Dx b. Dx b.39) +σx (t. s. T ] the mapping Rd → CB . X(t. x). x))(η h (t. (ii) We have (1) sup ([b(t. h ∈ Rd . s. x)dt (7. h dη (t. s.36) ([0. 7. s. x) = h. s.37) We set CB = CB ([s. x) · h = η h (t.1. x) Theorem 7. Rd )). x)) · η h (t. that Hypothesis 7. x). dB(t)) h η (s. L2 (Ω. 1]d . s. (1) Recall the notations given at the beginning of Chapter 6.1 holds.19 Assume that Hypothesis 7.T ] (7. s. x) with respect to x In this section we assume. Then for any s ∈ [0. x). x. besides Hypothesis 7. ·)]2 ) < ∞. 1) we have E X(t.Stochastic evolution equations Now from Proposition E. y ∈ [0. ·) belongs to C −d/(2m) (7. is continuously Gateaux diﬀerentiable and its Gateaux derivative is given by Xx (t. . ·)]2 + [σ(t. ·)2m < +∞.38) where η h (t. s.4. x) is the solution to the stochastic diﬀerential equation with random coeﬃcients. Dx σ and Dx σ are continuous on [0.20 Assume that Hypotheses 7. s. s. s. T ] × Rd . s. x) = bx (t.3 hold. X(t. let 0 ≤ s < t ≤ T and x. (7.1 and 7. 7.
To prove the theorem we use Theorem D.6. s. X.4.3 hold.k (t. t ∈ [s.2). setting Xxx (t. T1 ]. (the straightforward proof is left to the reader) and that for each x ∈ Rd . Y ∈ CB we have Fx (x. We set CB = CB ([s. s. X(r))Y (r)dB(r). t ∈ [s. that is F (x.41) Then F fulﬁlls Hypothesis D. x ∈ Rd . X(x)) = X(x). X)·Y ](t) = s bx (r. Note that the coeﬃcients of equation (7.39) fulﬁll Hypothesis 7. is twice diﬀerentiable with respect to x in any couple of directions (h. t t [FX (x. It is not diﬃcult to check that F is Gateaux continuously diﬀerentiable. T1 ]. x) We now prove the existence of the second derivative of X(t. x)(h. T1 ]) and deﬁne a mapping F : Rd × CB → CB . (7. X(r))dr + s σ(r. x). s. h ∈ Rd . x). 7. so it possesses a unique solution by Theorem 7.3. X(r))dB(r). which depends continuously on x.1 and 7. X(x) coincides with the solution X(·.40) where T1 > s is chosen such that F (x. So. Then the mapping Rd → CB . s.6 from Appendix D (with Λ = Rd and E = CB ). s.10. X)](t) : = x + s b(r. k) = ζ h.2 Existence of Xxx (t. x. x) of (7. x ∈ Rd . X(r))Y (r)dr+ s σx (r. Moreover. s.102 Chapter 7 Proof. 1 X1 − X2 2 CB for all X1 . x → X(·.21 Assume that Hypotheses 7.1 so that it possesses a unique ﬁxed point X(x) ∈ CB . Theorem 7. (7. X2 ) CB ≤ (7. k) in Rd . X) = I. x) with respect to x. X1 ) − F (x.42) . X2 ∈ CB . the conclusion follows from Theorem D. setting t t [F (x.
k d ζ (t. s. s. s. s. x)4 ≤ 27 + C1 s t η(r.k (t. x). s. x)) · (η h (t. η k (t. We ﬁrst prove a lemma. X(r. Lemma 7. s. s. Proof. s. x))η(r. s. T ]. x) ∈ CB ([s. s. s. X(t. x) = 1 + s t bx (r. X(r. x). X(r. x). x))η(r. s. x)dr (7. s.k ζ (s. s. By using (7. We have. s. s. s. X(t. (7. x)4 dr 4 +C1 s σx (r. η k (t. s. s. s.44) + s σx (r. x))(ζ h. x)dB(r). Then η(·. s. x))η(r. X(r.k (t. x) 4 ≤ 27 + 27 s t bx (r. s. t 4 ∀ s ∈ [0. x) = bx (t. x))(η h (t. s. s. dB(t)) h. x))η(r. x)4 ≤ C. x)dt +bxx (t. x ∈ Rd .37) and the H¨lder inequality we see that there exists a constant o C1 such that t η(t. s. x)dB(r) . x) = 0. x))η(r. x)dr 4 +27 s σx (r. T ). X(t. . x). x)dB(r) . s. x) is the solution to the stochastic diﬀerential equation (with random coeﬃcients) h.Stochastic evolution equations 103 ζ h. s. x) ∈ CB ([s.22 Let η(·. s. L4 (Ω)) and there exists C > 0 such that Eη(·. s. L2 (Ω)) be the solution of the equation t η(t. X(r. s.43) We shall prove the theorem when n = r = 1 for simplicity. T ].k (t. x))dt +σx (t. (7. X(t.45) η(t. x)) · ζ h. dB(s)) +σxx (t.
x))Z(r)η(·. x)dr (7. For any x ∈ R we deﬁne a linear bounded operator T (x) from CB into CB setting for all t ∈ [s. L (Ω)).8. x) = Xx (·.50) . s.21. s. s. We choose T1 as in (7. x)dB(r). T ]. by a straightforward computation ηx (·. T ]. x) = 1 + T (x)η(·.44). s. s. (7. T1 ]) as before. (7. By Theorem 7. s. L4 (Ω)) and it results t (T (x)Z)(t) = − s t bxx (r.104 Chapter 7 Now.41) it follows that T (x) L(CB ) (7. ∀ x ∈ R. s. s.20 we know that X(t. x) = (1 − T (x))−1 (T (x)η(·. x))Z(r)dr − s 4 σx (r. x)4 ≤ C2 (1 + s Eη(r. x). t t (T (x)Z)(t) = − s bx (r. s. s. x) belongs to CB and fulﬁlls equation (7. s. X(r. s. (7. The conclusion follows from the Gronwall lemma. s. since η(·. x))Z(r)η(·. x) By (7.49) From this identity it is easy to show the existence of ηx (·. r. X(r. x) := ζ(·. X(r. s. T (x)Z is diﬀerentiable with respect to x for any Z ∈ CB ([s. s. T1 ].48) is given by η(·. We have in fact. where C2 is another constant. X(r. x) is diﬀerentiable with respect to x and that its derivative η(·. Thus the solution of (7. Proof of Theorem 7. 0 ≤ s ≤ t ≤ T. x ∈ R.47) − s σxx (r.41) and CB = CB ([s.48) ≤ 1/2. x) = (1 − T (x))−1 (1). taking expectation on both sides of this inequality and using Corollary 6. x))Z(r)dB(r). Now we write equation (7. r. we ﬁnd that t Eη(t.44) as η(·. x)4 dr). x) ∈ CB ([s.46) Notice that. s. x)). s.
s. s. X(r. X(r. x) with reo spect to s. 7. s. s. x) is C 1 in all variables). s. (7. s. x)) · Xt (r. s. x)dB(r). s. x)(t) = s bxx (r. x))η 2 (·. X(r. r.51) t + s σxx (r. t ≥ r ≥ s.50) it follows that t ηx (t. x)dr (7. x) = X(t. x). X(r.1 and 7. s. s. x))η 2 (·. X(r. Let us compute Xs (t. s. x) − T (x)ηx (·. x))η 2 (·. s. X(r.53) with respect to r yields 0 = Xs (t. and the conclusion follows. s. 7.52). Setting r = s we ﬁnd Xs (t. x)) + Xx (t. X(r.3 with σ = 0.5 Itˆ Diﬀerentiability of X(t. x) the solution of (7. Now by (7.Stochastic evolution equations where t 105 T (x)η(·. x))η 2 (·. s. (7. T ].1 The deterministic case t ∈ [s. x)b(s. r. under Hypotheses 7. x). x)dB(r). x) = −Xx (t.53) Diﬀerentiating (7. s. s. x)). Denote by X(t. X(s) = x. X(t)). Write X(t. x)dr t + s σxx (r. s. . s. x) (it is well known that X(t. s.5.52) Let us consider the problem X (t) = b(t. r. x)(t) = s bxx (r. It is useful to recall ﬁrst some results in the deterministic case. s. s. s.
3. Let XN (t. 0 ≤ s ≤ t ≤ T. . A diﬃculty arises since the process s → X(t. s ∈ [0. s. x)du + s σ(u. where n ∈ N. Then B(t2 )−B(t1 ) and ϕ are independent.2). x)dB(u) = lim η→0 σ(tk−1 . Proposition 7. Proof. s. x) = x + s Xx (t. For this we need the following result which can be proved as Lemma 4. x)(B(tk ) − B(tk−1 )). N ∈ N. then X1 (t. Fs+ . s.54) In the next subsection we are going to generalize this formula for the solution X(t. r. however.24 Let t1 < t2 ≤ s.. x)dr. x) is not Fs measurable. that for any s ∈ [0.T ] is called the future ﬁltration of B. x) is Fs+ measurable. Let x ∈ Rd . and let ϕ ∈ L2 (Ω. because X(t. s. Since s t n σ(u.23 Assume that Hypotheses 7.. B(sn (ω)) − B(s(ω))) ∈ A} . x)b(r. x) be deﬁned by (7.106 which is equivalent to t Chapter 7 X(t. x) is Fs+ –measurable. .. P)..2 The stochastic case Here we want to study the diﬀerentiability of X(t. Then X(t. The family (Fs+ )s∈[0.. x)dB(u). s. Lemma 7. s. T ]. x) is not adapted. X(t. Then X1 (t. x) with respect to s in a sense to be precised. Now we introduce the backward Itˆ integral for a process wich is adapted o to the future ﬁltration.5. x) is measurable with respect to the σ–algebra Fs+ generated by all sets of the form {ω ∈ Ω : (B(s1 (ω)) − B(s(ω)). 0 ≤ s ≤ s1 < . < sn ≤ T and A ∈ B(Rn ). s.11). 7. x) = x + s b(u.1 holds. x) is Fs+ measurable. (7. s. x) of (7. s. We end the proof by recurrence. We have in fact t t X1 (t. s. s. It happens. k=1 where η = {s = t0 < t1 < · · · < tn = t}. T ].
(7. . r. L2 (Ω. x) − x = s Xx (t. Then we have X(t. Prove that t B(r)dB(r) = s 1 (B(t)2 − B(s)2 + (t − s)). L(Rr . T ]. σ(r. Rd ))) deﬁned in Chapter 5. 2 T (7. x) · b(r. L2 (Ω. x)(σ(r. (7.5. L(Rr . L2 (Ω.3 hold.25 For any F ∈ CB + ([0. For any η ∈ Σ with η = {0 = s0 < s1 < · · · < sn = T } we set n Iσ (F ) = k=1 F (tk )(B(tk ) − B(tk−1 )) The proof of next theorem is completely similar to that of equation (5. T ]. T ]. Rd ))) there exists the limit T σ→0 lim Iσ (F ) =: 0 F (s)dB(s). x).1 and 7.56) and T E 0 T 0 F (s)dB(s) = 0 E F (s) 2 HS ds. dB(r)) . s. L2 (Ω. r. Let F ∈ CB + ([0. x)(σ(r. x))]dr s (7.57) F (s)dB(s) is called the backward Itˆ integral of the function F in [0. L2 (Ω. T ]. Rd ))).58) + s Xx (t. x)dr t 1 + 2 t TR [Xxx (t.10).55) in L2 (Ω). L(Rr . L(Rr . T ]. x). T ]. Theorem 7. 2 7. r. Moreover we have T E 0 F (s)dB(s) = 0. o Exercise 7. Rd )) are called stochastic processes adapted to the future ﬁltration (Ft+ ) and continuous in quadratic mean. L(Rr .26 Let t > s.3 Backward Itˆ’s formula o t Theorem 7.27 Assume that Hypotheses 7. Rd ))) by a straightforward generalization of the space CB ([0.Stochastic evolution equations 107 We deﬁne CB + ([0. The elements of CB + ([0.
x)(x − X(sk . x) − x = − k=1 n [X(t.59) =− k=1 n Xx (t. x)(B(sk ) − B(sk−1 )) + o(sk − sk−1 ). sk−1 . We take d = r = 1 for simplicity. x)ek ) and (ek ) is any orthonormal basis in Rd . x)(x − X(sk . sk . sk−1 . k=1 Arguing as in the proof of Itˆ’s formula one can show. after some tedious but o straighforward computations. σ(r. t) we set η = max (tk − tk−1 ).n If η ∈ Σ(s. sk−1 . sk−1 . x) − X(t. t) we have n X(t. x))dB(r) = b(sk . x)(σ(r. sk−1 . k=1. x)(σ(r. that η→0 lim o(η) = 0. x) − x = sk−1 sk b(r. X(r. sk .. x))dr (7.60) + sk−1 σ(r. sk .. x). x))] = k=1 Xxx (t. X(r.. x)(sk − sk−1 ) + σ(sk . Proof. X(sk . sk−1 . sk . sk−1 .. x)] =− k=1 n [X(t.s. x)) − 1 2 Xxx (t.. x) − X(t. s. Pa. sk . . r. x)ek . σ(r.108 where d Chapter 7 TR [Xxx (t. r. For any η ∈ Σ(s. x))] (7. On the other hand we have sk X(sk . x))2 + o(η).
x)dB(r). x))]. x)σ(r. x) with b(ξk . . o 2 Theorem 7.59) we ﬁnd that n X(t. x)σ ∗ (r.61) n 1 + 2 Xxx (t. Obviously t η→0 lim I1 (η) = s Xx (r. x) dr 1 + 2 t t 2 Tr [Dx [ϕ(X(t. x)σ(sk . Concerning I2 (η). x)σ 2 (sk .62) + s Dx [ϕ(X(t.) Substituting (7.28 Let ϕ ∈ Cb (Rd ).60) in (7. x))]σ(r. x)]dr s (7. x)dB(r). sk . x)) − ϕ(x) = s Dx [ϕ(X(t. x))]. since b is deterministic. b(r. r. x) is Fs+ measurable by Proposition o k 7. x) where ξk is any point in [sk−1 . r. x)dr. σ(r. sk . s. In a similar way one can prove the following backward Itˆ formula. Then for any 0 ≤ s < t ≤ T. The other terms I3 (η) and o1 (η) can be handled as in the proof of Itˆ’s o formula. we have t ϕ(X(t. sk ]. x)b(sk . we note that it is an integral sum corresponding to the backward Itˆ integral since Xx (t.Stochastic evolution equations 109 (Notice that. r. sk .23. x)(B(sk ) − B(sk−1 )) (7. x)b(r. Therefore we have t η→0 lim I2 (η) = s Xx (r. s.60) b(sk . x) − x = k=1 n Xx (t. sk . one can replace in (7. x)(sk − sk−1 ) + k=1 Xx (t. x)(B(sk ) − B(sk−1 ))2 k=1 +I1 (η) + I2 (η) + I3 (η) + o1 (η).
110 Chapter 7 .
T ] × Rn .1) We consider here the problem X (t) = b(t. x ∈ Rn . under Hypothesis 8. t ∈ [0.t ϕ(x) = ϕ(X(t.2) with respect to u and setting u = s we ﬁnd Xs (t. s. s. (8. deﬁned on the space Cb (Rn ) by Ps. x)).3) Of great interest for the applications is the transition evolution operator Ps. x) + Xx (t. s. s. y ∈ Rn .1 The deterministic case t ∈ [s. (ii) There exists M > 0 such that b(t. x ∈ Rn .Chapter 8 Kolmogorov equations 8.1) has a unique solution X(·) = X(·. where s ∈ [0. T ]. t ∈ [0. (8. Rn ). u. X(s) = x ∈ Rn . X(u. 0 ≤ s ≤ u ≤ t ≤ T.1 (i) b is continuous on [0. x) − b(t.2) Morever. diﬀerentiating (8. T ]. T ]. s. T ] × Rn . x. (iii) b is diﬀerentiable with respect to x and bx is continuous on [0. s. As well known. (8. T ] × Rn → Rn fulﬁlls the following hypothesis.1 problem (8. T ) and b : [0. 0 ≤ s ≤ t ≤ T. Hypothesis 8. y) ≤ M x − y. s. 111 x ∈ Rn . x) · b(s. and it holds X(t. x) = 0. x) ∈ C 1 ([s. T ]. (8.4) . X(t)). t ∈ [0. T ]. x)). x) = X(t. s.t .
We have t≥s (8. x) + b(s. From (8. (s. x)) dt dt and Ps. X(t. Moreover for any ϕ ∈ Cb (Rn ) the mapping [0. x).9) z(T. ϕx (X(t.t ϕ = Ps. zx (s. Let us now consider the following partial diﬀerential equation called transport equation zs (s. Ps. ϕx (x) . T ] × [0.7) (8. s. s. T ] (8.t ϕ(x). x)).5) 1 Proposition 8. s. Let us prove (8.3). s.t ϕ = −L(s)Ps.t is a linear bounded operator on Cb (Rn ). (8. ds L(t)ϕ(x) = b(t. Xx (t.6) t ≥ s. x)). 1 where ϕ ∈ Cb (Rn ) and T > 0 is ﬁxed. T ]. . taking into acccount (8. x). d d Ps.t ϕ(x) = ϕ(X(t. x) = ϕ(x). is continuous.t L(t)ϕ. x)).7).t ϕ. T ] × Rn → Rn .t ϕ(x) = ϕ(X(t. x) = 0. x)) = b(t. x)) = − ϕx (X(t.t = Ps. Proof. dt and d Ps. t.6) follows. s. ϕx (X(t. s.8) where 1 ϕ ∈ Cb (Rn ). We have. u ∈ [0. s. x)) . s.u Pu. s ∈ [0.t L(t)ϕ(x) = b(t.t ϕ(x). x ∈ Rn . s.112 Kolmogorov equations As easily checked. x) · b(s.t .2) it follows immediately the cocycle property Ps.1 For any ϕ ∈ Cb (Rn ) we have d Ps. (8. x) → Ps. so that (8. d d Ps. t. x) ds ds = −L(s)Ps. X(t.
11). X(s. ϕx (x) . t ≥ 0. u. T ]. u.6). In this case it is easy to check that for any t > s ≥ 0. u.10). x)). x)).9) by (8.15) 1 ϕ ∈ Cb (Rn ). x ∈ Rn . By Proposition 8. X(s. x)) = z(u. X(s. x)) is constant in s. s.1. 8.9) we have d z(s. u. x) = b(x) and consider the problem X (t) = b(X(t)). we have Ps.5) it follows the semigroup law Pt+s = Pt Ps .t = P0.T ϕ(x) = ϕ(X(T. x) = Ps.1 and 1 let ϕ ∈ Cb (Rn ). Therefore z(s.2 Assume that b : [0. (8. x) = zt (s. X(u. x)) as required. s. u. x ∈ Rn . Then problem (8. Setting s = T and s = u we ﬁnd that z(T. t≥0 (8.9) has a unique solution z. u. X(s. Xt (s. Deﬁne Pt ϕ(x) = ϕ(X(t. t. u. x). s ≥ 0. x ∈ Rn . ϕ ∈ Cb (Rn ). given by (8. x)) = 0. u. n (8. T ] × Rn → Rn fulﬁlls Hypothesis 8. x)) ds = zt (s. x)) + zx (s. u. It is enough to notice that z.14) (8. X(s. x)) + zx (s. X(s.11) whose solution we denote by X(·. is a solution of (8.3 For any ϕ ∈ Cb (Rn ) we have Dt Pt ϕ = Pt Lϕ = LPt ϕ.10) Proof Existence. x)) which implies z(u. Uniqueness. x)).t−s . X(T. s ∈ [0. X(0) = x ∈ R .13) Pt is called the transition semigroup associated with (8. z is given by z(s. x) = ϕ(X(T. (8. b(s. u. x)). t ≥ 0. (8.Chapter 8 113 Theorem 8.12) so that by (8. X(s. If z is a solution of problem (8.1 we deduce 1 Proposition 8. .1 The autonomous case We assume here that b(t. where Lϕ(x) = b(x).
t . t. x)−b(t.2 we have Kolmogorov equations 1 1 Theorem 8. x) = ϕ(x).17) 8. Ps. x ∈ Rn . is continuous for all ϕ ∈ Cb (Rn ). s. Ps. by Theorem 8.18). t ≥ 0. . x) = Pt ϕ(x) = ϕ(X(t.t ϕ(x) = E[ϕ(X(t. s with 0 ≤ s ≤ t ≤ T and for all function ϕ ∈ Cb (Rn ) we set Ps. T ] × Rn . Then problem ut (t. For all t. s.18) and assume that the following hypothesis holds. has a unique solution given by u(t. t ≥ 0. 0 ≤ s ≤ t ≤ T. ux (t. By Chapter 6 we know that the mapping (s. (iii) b and σ have ﬁrst and second partial derivatives with respect to x continuous and bounded in [0. X(t))dB(t) X(s) = x ∈ Rd (8. x) the solution of (8.t ϕ(x). x.t is a linear bounded operator on Cb (Rn ). is called the transition evolution operator associated with (8. (8. T ] × Rn → L(Rr . T ] × Rn → Rn and σ : [0. x ∈ Rn (8.4 Assume that b ∈ Cb (Rn ) and let ϕ ∈ Cb (R). x)−σ(t. Hypothesis 8. 0 < s ≤ t ≤ T. x) → Ps. Rn ) are continuous. (ii) There exists M > 0 such that b(t.19) As easily checked. y)+ σ(t. We denote as before by X(·.2 (i) b : [0.2 Stochastic case We consider the stochastic evolution equation dX(t) = b(t. t ∈ [0. x ∈ Rn . y) HS ≤ M x−y. x)).18) corresponding to η = x ∈ Rn . y ∈ Rn . x))].114 Finally. X(t))dt + σ(t. T ].16) u(0. (8. x) . x) = b(x). x ∈ Rn .
x))]dr. 2 Proposition 8. s. that is Ps.2 holds and let ϕ ∈ Cb (Rn ). s. x).t ϕ is diﬀerentiable in s and we have d Ps. dt Proof. Then Ps. s. ds t ≥ 0.20) The ﬁrst basic identity is the following.21).t L(t)ϕ.3 Basic properties of transition operators 1 Tr [ϕxx (x)σ(s. 2 Let us introduce the Kolmogorov operator (L(s)ϕ)(x) = 2 ϕ ∈ Cb (Rn ). The second basic identity is the following.5 Assume that Hypothesis 8. (8.t ϕ is diﬀerentiable in t and we have d Ps. ϕx (x) . .21) dt ϕ(X(t. (8. σ(t. which yields (8. Taking expectation in the backward Itˆ formula (7.t ϕ = −L(s)Ps. x))dB(t) . yields t E[ϕ(X(t. s. s. x)σ ∗ (s.t ϕ.t (L(r)ϕ)(x)dr.t ϕ(x) = ϕ(x) + s t Pr.6 Assume that Hypothesis 8. x)) = (L(t)ϕ)(X(t.t ϕ = Ps. Integrating with respect to t and taking expectation. X(t.22). which coincides with (8. By the Itˆ formula we have that o t ≥ 0.2 holds and let ϕ ∈ Cb (Rn ). Then Ps. x))] = ϕ(x) + s E[(L(r)ϕ)(X(r. x)] + b(s. x)) + ϕx (X(t.Chapter 8 115 8.22) Proof.62) we ﬁnd o t Ps. (8. x)).r ϕ(x)dr.t ϕ(x) − ϕ(x) = s L(r)Ps. 2 Proposition 8. s.
ϕ ∈ Cb (Rn ). taking expectation we ﬁnd z(u.116 Kolmogorov equations 8. u. Integrating in s between u and T yields z(T. since z fulﬁlls (8. x) = ϕ(x). T ] × Rn → R is a solution to (8. z is given by z(s. x))dB(s) . x))].t = Ps.r Pr. X(s. u. u. x)) = zs (s. x)) − z(u. By (8. (8. u. σ(s. X(s. x) = E[ϕ(X(t. x) = E[ϕ(X(T. and let 0 ≤ u ≤ s ≤ T. σ(s. X(s. x)) = ϕ(X(t.23). X(s.7 Assume that Hypothesis 8. 2 0 < s ≤ T.8 Prove the cocycle law Ps. fulﬁlls (8. zx .24) Proof.23) We consider here the parabolic equation zs (s. s. X(T.2 holds and let ϕ ∈ Cb (Rn ). ·)))(x) + zx (s. T ]. x)). 2 Theorem 8. (8. u. x)) − z(u. n We say that a function z : [0. X(u. u. and fulﬁlls (8. x)). u. x))dB(s) = zx (s. u.25) . u. We have o ds z(s. x ∈ Rn .22) it follows that z(s. x))].23). Then there exists a unique solution z of problem (8. (8. x) = Ps. X(s. x) t s ∈ [0.23). u. Existence. u. u. x) + (L(s)(z(s. X(s. X(s. X(s. Now. x))dB(s).T ϕ(x).23). x ∈ R . ·)))(x) = 0.23) if z is continuous and bounded together with its partial derivatives zt . = u zx (s. x))ds + (L(s)z(s. z(T. u.4 Parabolic equations 0 ≤ s < T. Let z be a solution to (8. Exercise 8. X(s. X(s. u.23). Let us compute the Itˆ diﬀerential of z(s.t for 0 ≤ s ≤ t ≤ t ≤ T. x))σ(s. Uniqueness. x)). zxx .
s. x). x) be the solution of the stochastic evolution equation dX(t) = b(X(t))dt + σ(X(t))dB(t) (8. x). x ∈ Rn .7 we ﬁnd the result . Set Y (t) = X(t + a. problem (8. x) = u(t. x). Proof. s ≥ 0. s. v(0. 2 Proposition 8. The we have t+a t+a X(t+a. s ∈ [0. t]. x) and X(t + a. t].Chapter 8 117 8. x) coincide. Assume that b and σ are independent of t : Then we have L(s) = L where 1 2 Tr [ϕxx (x)σ(x)σ ∗ (x)] + b(x). t. t ≥ 0. Now the conclusion follows. s+a.4. x) = x+ s+a b(X(r. x) = Lv(s.26) but with the Brownian motion B(t) replaced by B1 (t). setting Pt = P0. ϕx (x) .26) X(s) = x ∈ Rn . σ(t. x ∈ Rn . Setting v(s. t ≥ 0 is a semgroup of linear operators in Cb (Rd ). s + a. ϕ ∈ Cb (Rn ). we have Pt+s = Pt Ps .27) Then by Theorem 8. Setting r − a = ρ yields t t Y (t) = x + s b(Y (ρ))dρ + s σ(Y (ρ))d[B(ρ + a) − B(a)]. s + a.1 Autonomous case b(t. t ≥ 0. x) = b(x). x ∈ Rn . s+a. x))dB(r). Setting B1 (t) = B(t + a) − B(a) we see that Y (t) fulﬁlls equation (8. s+a. Thus Pt . P0 = 1. s ∈ [0. x) = ϕ(x). By the proposition and the cocycle law (8.t . x) = σ(x). t − s. x))dr + s+a σ(X(r.9 Let X(t.25)it follows that. Lϕ(x) = Then for any and a > 0 the laws of X(t.23) becomes vs (s. x ∈ R (8.
The solution of (8.27) has a unique solution given by v(s. G .28) 8. (8. P) taking values in Rn .29) where A.10 Assume that b. x) = ϕ(x).t ϕ(x) = Pt ϕ(x).Qt (dy). x) = Pt−s.30) is given by the variation of constants formula t X(t.32) esA QesA ds.30) where B is a standard Brownian motion in a probability space (Ω.34) So. . σ : R → R are Lipschitz continuous and of 2 class C 2 . x)# P = NetA x.29) is given by u(t.Qt . Then. The corresponding stochastic diﬀerential equation is √ dX(t) = AX(t)dt + Q dB(t). x)] + Ax + ux (t. Q ∈ L(Rn ).31) Therefore the law of X(t. ∗ t ≥ 0. X(0) = x. x) is given by X(t.33) where A∗ is the adjoint of A.11 Consider the parabolic equation in Rn 1 ut (t.118 Kolmogorov equations Theorem 8. (8. x) = 2 Tr [Quxx (t. (8. (8. x) u(0. problem (8. s ∈ [0. t ≥ 0. x ∈ R. the solution of (8. where Qt = 0 t (8.5 Examples Example 8. Consequently. t]. the transition semigroup Pt looks like Pt ϕ(x) = Rn ϕ(y)NetA x. x) = Pt ϕ(x). x) = e x + 0 tA e(t−s)A QdB(s). (8. x ≥ 0 for all x ∈ Rn . Q is symmetric and Qx. (8. for any ϕ ∈ Cb (R).
P). The solution of (8. F .37) is given by X(t. det Qt > 0 we have u(t. x) = ϕ(x). (8. in particular.37) where B is a real Brownian motion in is a real Brownian motion in some probability space (Ω.Chapter 8 If. x) + axux (t.38) e− 2t ϕ(e(a−q/2)t+ y2 √ qy x)dy. x) = e(a−q/2)t+ Therefore 1 Pt ϕ(x) = √ 2πt +∞ −∞ √ q B(t) x. X(0) = x.36) where q > 0 and a ∈ R.35) Example 8. (8. x) = 1 qx2 uxx (t.39) . (8. (8. The corresponding stochastic diﬀerential equation is √ dX(t) = aX(t)dt + q X(t)dB(t).12 Consider the parabolic equation in R ut (t. x) = (2π)−n/2 [det Qt ]−1/2 Rn 119 e− 2 1 Q−1 (y−etA x).(y−etA x) t ϕ(y)dy. (8. x) 2 u(0.
120 Kolmogorov equations .
Theorem A.Appendix A λsystems and πsystems Let Ω be a non empty set. a λsystem if (i) Ω. Proof. A non empty family R of parts of Ω is called a πsystem if A. Let D0 be the minimal λsystem including R. D ⊂ σ(R) we have σ(R) = D. For this it is enough to show. ∅ ∈ D. Then we have σ(R) ⊂ D.1) ∞ i=1 Ai ∈ D. 121 (A. B ∈ R =⇒ A ∩ B ∈ R.2) . which will imply the theorem. where σ(R) is the σ algebra generated by R. (ii) A ∈ D =⇒ Ac ∈ D. Moreover. If in particular. if D is a λsystem such that A. We are going to show that D0 is a σ–algebra. (iii) (Ai ) ⊂ D mutually disjoint =⇒ (A. B ∈ D =⇒ A ∩ B ∈ D then it is σ–algebra. i=1 Let us prove the following Dynkin theorem. as remarked before. In fact if (Ai ) is a sequence in D of not necessarily disjoint sets we have ∞ Ai = A1 ∪ (A2 \ A1 ) ∪ (A3 \ A2 \ A1 ) ∪ · · · ∈ D i=1 and so ∞ Ai ∈ D by (ii) and (iii). that the following inclusion holds A.1 Let R be a πsystem and let D be a λsystem including R. B ∈ D0 =⇒ A ∩ B ∈ D0 . Obviously any algebra is a πsystem.
2 Let A be an algebra of subsets of Ω and let F be the σalgebra generated by A .3) is fulﬁlled. It is clear in fact that A is a πsystem. B ∈ D0 ⇒ R ∩ B ∈ D0 .1 it follows that P1 = P2 . that F ∪ B c ∈ D0 . ∀ I ∈ A . If we show that H (B) ⊃ R. we have that F ∪ B c ∈ D0 as required. since F ∪ B c = (F \ B c ) ∪ B c = (F ∩ B) ∪ B c and F ∩ B and B c are disjoint. F ) such that P1 (I) = P2 (I). equivalently. the following implication holds R ∈ R. . Using the Dynkin theorem we can show that P1 = P2 .122 For any B ∈ D0 we set λsystems and πsystems H (B) = {F ∈ D0 : B ∩ F ∈ D0 }. So. ∀B ∈ D0 (A. Let P1 and P2 be probability measures on (Ω. It is easy to see that D is a λsystem which contains D.2) is proved. Therefore H (R) = D0 by the minimality of D0 . In fact properties (i) and (iii) are clear. Example A. Deﬁne D = {B ∈ F : P1 (B) = P2 (B)}. Consequently. On the other hand it is clear that if R ∈ R we have R ⊂ H (R) since R is a πsystem.3) then we conclude that H (B) = D0 by the minimality of D0 and (A. In fact. which yields R ⊂ H (B) and (A. We claim that H (B) is a λsystem. by Corollary A. It remains to show that if F ∩ B ∈ D0 then F c ∩ B ∈ D0 or.
P) of L2 (Ω. Let us consider the signed measure µ(G) = G XdP. G . P) and a σalgebra G included in F . F . by the RadonNikodym Theorem there exists a unique Y ∈ L1 (Ω. ∀ G ∈ G. F . P) (1) . We say that X is G measurable if I ∈ B(R) ⇒ X −1 (I) ∈ F . P). F . (B. P) such that µ(G) = G XdP = G Y dP.Appendix B Conditional expectation B. It is clear that µ is absolutely continuous with respect to the restriction of P to G . Therefore. (B. Show that E(XG ) coincides with the orthogonal projection of X into the closed subspace L2 (Ω.1 Deﬁnition We are given a probability space (Ω. F . It is clear that X is not G measurable in general. G ∈ G. (1) 123 .1 Assume that X ∈ L2 (Ω. it is denoted by E(XG ).1) E(XG ) is characterized by XdP = G G E(XG )dP. ∀ G ∈ G.1) The G measurable random variable Y is called the conditional expectation of X given G . Let X : Ω → R be a real random variable on (Ω. In view of (B. G . In all this appendix by random variable we mean an equivalence class of random variables with respect to the usual equivalence relation.2) Exercise B. P).
s. Assume that X is G measurable. Then we have XdP = A A (B.3) Moreover. (B. P). we have E(XG ) = X. Y.9) So. Proposition B. Then we have E(XY G ) = XE(Y G ). F . E(αX + βY G ) = αE(XG ) + βE(Y G ). (B. Let A ∈ H . one can check easily the linearity of conditional expectation.6) 1 A XdP = P(A)E(X) = l A E(XG )dP.9) we see that E(XH )dP = A A XdP = A E E(XG ) H dP. comparing (B.8) and XdP = A A E(XG )dP = A E E(XG ) H dP.10) . Y ∈ L1 (Ω. Pa.4 Let X. β ∈ R and all X. P). Pa. P) and let G be σalgebra included in F .2) yields E[E(XG )] = E(X).2 Basic properties Let X. Y ∈ L1 (Ω.8) and (B. Also if X ≥ 0.4) for all α. one has E(XG ) ≥ 0. Proof. Proposition B.3 Let H be a σalgebra included in G ..s. Then we have E(XG ) = E(X). It is obvious that if X is G measurable. Then 1 A and X are independent so that l XdP = A Ω (B. Then we have E(XH ) = E E(XG ) H . Let A ∈ G .124 Conditional expectation B. From this one deduces the inequality E(XG ) ≤ E(X G ). (B. Setting G = Ω in (B.7) E(XH )dP (B.5) (B. XY ∈ L1 (Ω. Proof. F . F . (B.2 Assume that X is independent of G . Proposition B.
Y )dP = G G (B.5 Let X. P) and let φ : R2 → R be bounded and Borel. P). dy. then since G ∩ A ∈ G we have E(1 A Y G )dP = l G G 1 A Y dP = l G∩A Y dP = G∩A E(Y G )dP = G 1 A E(Y G )dP. This is clearly equivalent to E(Zφ(X. Proof. F . Z)# P. Z) with values in R3 µ = (X. Y )G ) = h(X). (B. Let us prove now a useful generalization of this Corollary. Assume that X is G measurable and Y is independent of G .15) .6 Let X. Y ∈ L1 (Ω. Then we have E(φ(X. (B. It is enough to show (B. Corollary B. Recalling Proposition B. ∀Z ∈ L1 (Ω. Y )].Appendix B 125 Proof.10) for X = 1lA where A ∈ G . Y. F . XY ∈ L1 (Ω. (B. dz). So. Y )) = R3 zφ(x.13) h(X)dP. Proposition B.2 we ﬁnd. Let now G ∈ G .14) Denote by µ the law of the random variable (X. Then we have E(XY G ) = XE(Y ). Y. Y )) = E(Zh(X)). P). We have to show that φ(X.11) x ∈ R. Assume that X is G measurable and that Y is independent of G . Y. y)µ(dx.12) where h(x) = E[φ(x. ∀ G ∈ G. (B. E(Zφ(X. G . l for any G ∈ G .
F . P). Therefore we can write (B.7 Let F. y)λ(dy) ν(dx. Exercise B. P) and Z = E(HG ). Y )) = R3 λ(dy) = Y# P(dy). dz)λ(dy). zφ(x. H. where ν(dx. y)ν(dx. dz). (B. Using the Fubini Theorem we get ﬁnally E(Zφ(X. Prove that E(F H) = E(F Z). dy. Prove the Jensen inequality E(g(F )G ) ≥ g(E(F G )).17) . dz) = (X.8 Let g : R → R be convex and let F. dz) = R2 zh(x)ν(dx. as required. Y )) = R2 z R φ(x. Z) and Y are independent so that µ(dx.126 Conditional expectation Since X and Z are G measurable and Y is independent of G . dz) = E(Zh(X)). G .15) as E(Zφ(X. (B. Z)# P(dx. dz)λ(dy). g(F ) ∈ L1 (Ω.16) Exercise B. F H ∈ L1 (Ω. the random variables (X. dz) = ν(dx.
A ∈ Fs .T ] is said to be a martingale (with respect to the ﬁltration (Ft )t≥0 ) if E[M (t)Fs ] = M (s). a supermartingale if E[M (t)Fs ] ≤ M (s). (Ft )t≥0 an increasing family of σalgebras included in F and (M (t))t∈[0. Thus (M (t))t∈[0. T ]. ∀ 0 ≤ s < t ≤ T. ∀ 0 ≤ s < t ≤ T. ∀ 0 ≤ s < t ≤ T.T ] is a martingale if and only if M (s)dP = A A M (t)dP. and a supermartingale if and only if M (s)dP ≤ A A M (t)dP. a stochastic process. a submartingale if E[M (t)Fs ] ≥ M (s). A ∈ Fs . ∀ 0 ≤ s < t ≤ T.Appendix C Martingales C.1 Deﬁnitions Let (Ω. A ∈ Fs . P). a submartingale if and only if M (s)dP ≥ A A M (t)dP. P) be a probability space. F . ∀ 0 ≤ s < t ≤ T. t ∈ [0.T ] with M (t) ∈ L1 (Ω. 127 . (M (t))t∈[0. ∀ 0 ≤ s < t ≤ T. Ft .
4 For all λ > 0 we have P(S ≥ λ) ≤ 1 λ M (tn )dP.2 The basic inequality for martingales S = sup M (ti ). A− = {ω ∈ Ω : M (s)(ω) ≤ 0}. C. Example C. A ∈ Fs . (See Exercise B. Proof. Proposition C. Exercise C. let 0 < t1 < t2 < . Clearly A+ and A− belong to Fs .1) . This shows that M  is a submartingale. In fact.8). 1≤i≤n Let M (t) be a martingale.3 Using Jensen’s inequality prove that any convex function of a martingale is a submartingale. {S≥λ} (C. Set A+ = {ω ∈ Ω : M (s)(ω) > 0}. Since B(t) − B(s) and 1 A are independent we have l (B(t) − B(s))dP = E(1 A (B(t) − B(s))) = 0. < tn ≤ T and set We are going to prove an important estimate (due to Kolmogorov) of S in terms of M (tn ).1 If M is a martingale then M  is a submartingale..2 The Brownian motion B is a martingale. l A so that B(t)dP = A A B(s)dP. let t > s and A ∈ Fs . Consequently we have M (s)dP = A A+ M (s)dP − A− M (s)dP = A+ M (t)dP − A− M (t)dP ≤ A M (t)dP..128 Martingales Proposition C. Let 0 ≤ s < t ≤ T.
129 Clearly. P) for all t ∈ [0.... n. T ]. . Let us estimate {S≥λ} M (tn )dP. i = 1. recalling that M (t) is a sub– λP(An−1 ) ≤ An−1 M (tn−1 )dP ≤ An−1 M (tn )dP.Appendix C Proof. (C... · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·· An = {M (t1 ) < λ. F . . . We have obviously M (tn )dP ≥ λP(An ).. sets A1 . M (t2 ) ≥ λ}.2) Summing up on k from 1 to n the conclusion follows. Set A1 = {M (t1 ) ≥ λ}.. 1≤i≤n We are going to estimate of E[S 2 ] in terms of E[M 2 (tn )].. We have. .. and we have n {S ≥ λ} = i=1 Ai . Therefore M (tn )dP ≥ λP(An−1 ). Ak k = 1.. An are mutually disjoint. A2 = {M (t1 ) < λ. Let 0 < t1 < t2 < . < tn ≤ T and set as before S = sup M (ti ). An−1 X(tn )dP. An Now we estimate martingale. . C. n. Moreover Ai ∈ Fti . M (tn ) ≥ λ}.3 Square integrable martingales In this section we are given a martingale M (t) such that M (t) ∈ L2 (Ω. . . An−1 Proceeding in a similar way we obtain M (tn )dP ≥ λP(Ak ).. .
1) we have F (t) ≤ Consequently ∞ ∞ Martingales sup M (ti )2 1≤i≤n ≤ 4E(M (tn )2 ).+∞)×Ω 1 √ M (tn )1 {S≥√t} P(dω)dt l t ∞ = Ω M (tn )P(dω) 0 S2 1 √ 1 {S≥√t} dt l t 1 √ dt t 1/2 1/2 = Ω M (tn )P(dω) 0 =2 Ω M (tn )SP(dω) ≤ 2 Ω M (tn )2 dP Ω S 2 dP . Corollary C. t∈[0.5) . Set F (t) = P(S > t). {S≥t} (C. (C.T ] E (C. Then for any T > 0 we have sup M (t)2 ≤ 4E[M 2 (T )]. by (C.1) and the Fubini Theorem we have ∞ E(S ) ≤ 0 2 1 √ t √ {S≥ t} M (tn )dP dt = [0.130 Proposition C.5 We have E Proof.3) t ≥ 0.6 Let M be a square integrable continuous martingale. By (C. Now the conclusion follows easily. 1 t M (tn )dP. So.4) E(S 2 ) = 0 P(S 2 > t)dt = 0 P(S > √ t)dt.
Appendix C 131 Proof. that E as required. 1≤i≤m Since M is continuous it follows. By Proposition C. sup M (s)2 ≤ 4E M (T )2 . . Let 0 < s1 < s2 < · · · < sm = T. . sm . by the arbitrariness of the sequence s1 . s∈[0.T ] .5 it follows that E sup M (si )2 ≤ 4E M (T )2 . . s2 . .
132 Martingales .
y ∈ E. Theorem D. ∀ λ ∈ Λ. x) − F (λ. x. ∀ λ ∈ Λ. then x is of class C 1 and x (λ) = Fλ (λ. E be Banach spaces (norms  · ). We are given a continuous mapping The following result (contraction principle) is classical.1) (ii). a 133 . There exists a unique continuous mapping x : Λ → E. x(λ)). (λ. Let Λ. (D. We want to generalize the second part of this result to mappings F (λ. x(λ)) + Fx (λ. x) and assume that Hypothesis D. x) which are only continuously Gˆteaux diﬀerentiable. y) ≤ κx − y. x) → F (λ. x(λ))x (λ).2) λ → x(λ). (D. 1) such that F (λ. If in addition F is of class C 1 .1 There exists κ ∈ [0.Appendix D Fixed points depending on parameters D.1 Introduction F : Λ × E → E.1 (i). such that x(λ) = F (λ.
a → DΦ(a).4 Let A. (1) Example D. Proposition D. ∀ x. a Then the following identity holds 1 Φ(c) − Φ(a) = 0 DΦ((1 − ξ)a + ξc)(c − a)dξ. (as one can see) Φ is not diﬀerentiable in any point. Then we have F (ξ) = DΦ((1 − ξ)a + ξc)(c − a)dξ. If in addition for all c ∈ A the mapping A → B. c ∈ A. Deﬁnition D. B = L2 (0. Then one can check easily that Φ is continuously Gˆteaux diﬀerentiable and a DΦ(x)y = y cos x. ξ ∀ a.3 It is well known that if the mapping A → L(A. Set F (ξ) = Φ((1 − ξ)a + ξc). e . B). B). However. y ∈ L2 (0. One also says that Φ is Fr´chet diﬀerentiable.134 Fixed points D. 1) and Φ(x) = sin x.5 Let Φ : A → B be continuously Gˆteaux diﬀerentiable. a → DΦ(a)c is continuous we say that Φ is continuously Gˆteaux diﬀerentiable. 1]. We shall need the following result. a → DΦ(a) is continuous then Φ is diﬀerentiable.2 We say that Φ is Gˆteaux diﬀerentiable if there exists a a mapping DΦ : A → L(A. and the conclusion follows just integrating this identity between 0 and 1.3) Proof. (1) ξ ∈ [0.2 Gˆteaux diﬀerentiable mappings a Let A and B be Banach spaces and let Φ : A → B be a continuous mapping from A into B. 1). (D. a Remark D. such that ξ→0 lim 1 (Φ(a + ξc) − Φ(a)) = DΦ(a)c.
Theorem D. µ. x(λ + hµ)) − F (λ. x(λ) + ξ(x(λ + hµ) − x(λ))) · (x(λ + hµ) − x(λ))dξ.5) (D. ∀ λ ∈ Λ. x(λ)) · µ + Fx (λ. We assume that Hypothesis D. such that x(λ) = F (λ. z ∈ E. x(λ)+ξ(x(λ+hµ)−x(λ)))·zdξ. x(λ)) · µ. (D.Appendix D 135 D.7) Set now 1 G(λ. h)z = Gz := 0 Fx (λ+ξhµ. x(λ)))−1 Fλ (λ. Let λ. x) → F (λ.1 Gz ≤ κz.6 Assume that Hypotheses D. equivalently x (λ) · µ = Fλ (λ. ∀ z ∈ E. . x(λ))(x (λ) · µ). µ ∈ Λ and h ∈ R. (D.3) it follows that x(λ + hµ) − x(λ) = F (λ + hµ.4) and (D. x.1 is fulﬁlled and denote by x the mapping x : Λ → E. x). Then x(·) is continuously Gˆteaux diﬀerena a tiable as well and we have x (λ) · µ = (1 − Fx (λ. x(λ)).6) =h 0 1 Fλ (λ + ξhµ.4) λ → x(λ). Then G ∈ L(E) and by Hypothesis D. From (D. Proof. (λ.1 is fulﬁlled and that F is continuously Gˆteaux diﬀerentiable. x(λ)) 1 (D.3 The main result We can back to the notations of the introduction and consider two Banach spaces Λ and E and a continuous mapping F : Λ × E → E. x(λ) + ξ(x(λ + hµ) − x(λ))) · µdξ + 0 Fx (λ + ξhµ.
7) we have (1 − G(λ. Therefore x (λ) · µ − Fx (λ. x. h))−1 h 1 × 0 Fλ (λ + ξhµ. Letting h → 0 we ﬁnd x (λ) · µ = (1 − Fx (λ. x(λ)). which implies 1 x(λ + hµ) − x(λ)) = (1 − G(λ. x(λ)))−1 Fλ (λ. x(λ)). µ. µ. x. h))(x(λ + hµ) − x(λ)) 1 Fixed points =h 0 Fλ (λ + ξhµ.136 Then from equation (D. x(λ) + ξ(x(λ + hµ) − x(λ))) · µdξ. . x(λ))(x (λ) · µ) = Fλ (λ. x(λ) + ξ(x(λ + hµ) − x(λ))) · µdξ.
< Theorem E.2 (The Brownian motion) Let > 0 and let p ≥ 1. 1] 2m .T ]2 f (t) − f (s)2m dt ds t − s1+2m . T ]). then B 2m W .1) Example E. (E. Let us compute E( B p W .1 (Sobolev embedding) Assume that > 1/(2m).2m W .1 Fractional Sobolev spaces on [0.Appendix E Fractional Sobolev spaces and regularity of processes E. Then the following inclusion holds with continuous embedding.T ]2 t − sm dt ds = cm t − s1+2m 137 . T ) ⊂ C −1/(2m) ([0. W .T ]2 B(t) − B(s)p dt ds t − s1+p Take for simplicity p = 2m. T ] → R such that f +∞.2m (0.2m E =E [0. 1).p (0.T ]2 = cm [0. Deﬁne f := [0. T ) is by deﬁnition the space of all f : [0. T ) or not.2m (0.T ]2 B(t) − B(s)2m dt ds t − s1+2m t − sm−1−2m dt ds [0.p ) = E [0. m ∈ N. We ask the question whether B(·) belongs to W .2m Let ∈ (0.
E. and cm > 0 E[X(t) − X(s)2m ] ≤ cm t − sm . as the next proposition shows. T ) for < 2 . Assume that there is a > 0. s ∈ [0. 4 2 Arguing similarly taking larger m we conclude that B(·) ∈ C α (0. 1/2). Then we have E X2m < +∞.138 Fractional Sobolev spaces The integral is ﬁnite if and only if < 1 . Therefore 1 if 1 < < 1 we conclude by the Sobolev embedding that B(·) ∈ C − 4 (0. t ∈ [0. F . T ]. T ].3 Assume that there is m > 1. T ) Let (Ω. 1/2). (E.2m (0. b > 0 such that E[X(t) − X(s)1+a ] ≤ cm t − s1+b ∀ t. and cm > 0 (E. The last statement follows from the Sobolev embedding theorem. be a real stochastic process on (Ω. This does not imply that B(·) is continuous. P). T ) again for < 2 . ≤ cm [0. ∈ (0.2 (0. . T ]. o Proposition E. 1/2). ∀ t. such that (E. a (E. 2 1 For instance taking m = 1 we conclude that B(·) ∈ W .2) This estimate (provided m > 1) allows us to conclude that trajectories of X are H¨lder continuous almost surely.2 Processes belonging to W . 1 But if we take m = 2 we have B(·) ∈ W .T ]2 t − sm−1−2m dt ds < ∞. T ). s ∈ [0.2) is fulﬁlled. We have in fact E X 2m .4 Kolomogorov test It is a generalization Proposition E. T ) for any α ∈ (0. since ∈ (0. ω) belongs to C Proof.4 (0.3) ([0. One situation often encountered is when the following estimate holds for some m > 1. 1+b . 1/2) and m − 1 − 2m > −1.4) Then X has αH¨lder continuous trajectories with α < o . X(·.3. T ]) for almost ω ∈ Ω. Remark E. P) be probability space and let X(t).2m −1/(2m) ∈ (0. F .2m Moreover.
1).2m := [0. and cm > 0 such (E. F .6) This estimate implies that almost all trajectories of X are H¨lder continuous o almost surely. Then we have E X2m < +∞. Assume that there is m > 1.5) Let (Ω. 1/2) and m − 1 − 2m > −1. d ∈ N.6 Assume that there is m > 1. Theorem E. (E. T ].3 Multi dimensional Sobolev spaces and regularity of random ﬁelds f (x) − f (y)2m dx dy. since ∈ (0. X(·. 1). T ]) for almost ω ∈ Ω. T ]d ) ⊂ C −d/(2m) ([0.5 (Sobolev embedding) Assume that > d/(2m). x ∈ [0. and cm > 0 E[X(x) − X(y)2m ] ≤ cm t − s2m . W . Proposition E. 1).2m 2m . ∀ t.2m ([0. m ∈ N.2) is fulﬁlled. F . s ∈ [0.T ]2 t − sm−1−2m dt ds < ∞.7) ([0. (E.2m ) −d/(2m) ∈ (0. x − yd+2m Let ∈ (0. that (E. T ]d .T ]2d ([0. P) be probability space and let X(x). We have in fact E( X 2m . P). Then the following inclusion holds with continuous embedding. The last statement follows from the Sobolev embedding theorem. Deﬁne f W f . T ]d → R such that . ω) belongs to C Proof. ∈ (0. ≤ cm [0. .2m < +∞. .2m Moreover. T ]d ) is by deﬁnition the space of all f : [0.Appendix F 139 E. T ]d ). be a random ﬁeld on (Ω.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.