Øksendal: Stochastic Differential Equations

Solutions Manual
Christopher Kennedy

May 3, 2021
Contents

1 Introduction 2

2 Some Mathematical Preliminaries 3

3 Itô Integrals 9

4 The Itô Formula 14

5 Stochastic Differential Equations 20

6 The Filtering Problem 28

7 Diffusions: Basic Properties 29

8 Other Topics in Diffusion Theory 30

9 Applications to Boundary Value Problems 31

10 Applications to Optimal Stopping 32

11 Applications to Stochastic Control 33

12 Applications to Mathematical Finance 34

1
Chapter 1

Introduction

This is a solutions manual for Stochastic Differential Equations by Bernt Øksendal. This is a
working document last updated May 3, 2021. Progress to date:

• Chapter 2: Problems #1-17

• Chapter 3: Problems #1-17

• Chapter 4: Problems #1-15

• Chapter 5: Problems #1-17

• Chapters 6–12: none so far

2
Chapter 2

Some Mathematical Preliminaries

1. Suppose X : Ω → R is a function that assumes countably many values {aj } in R.

(a) Note that X is a random variable if and only if it is measurable. If X : Ω → R is

measurable, then U = X −1 (R \ ak ) ∈ F and thus X −1 (ak ) = Ω \ US∈ F, ∀k. On the
other hand, if X −1 (ak ) ∈ F, ∀k, then Borel set V ⊆ R, X −1 (V ) = ak ∈V X −1 (ak ) ∈
F and thus X is measurable.
(b) Compute E(|X|) = R |x| dPX = S∞ {ak } |x| dPX = ∞
R R P
k=1
k=1 |ak |P(X = ak ).

(c) If E(|X|) < ∞, then the series

Z Z ∞
X
E(X) = x dPX = S x dPX = ak P(X = ak )
∞
R k=1 {ak } k=1

is absolutely convergent and therefore converges.

(d) If f is measurable and |f | is bounded by M , then
Z Z Z
E(|f (X)|) = |f (x)| dPX ≤ M dPX = M dPX = M < ∞.
R R R

Hence,
Z Z ∞
X
E(f (X)) = f (x) dPX = S∞
f (x) dPX = f (ak )P(X = ak )
R k=1 {ak } k=1

is absolutely convergent and therefore converges.

2. Let F (x) = P(X ≤ x) be the distribution function of X.

(a) By monotonicity of P, 0 = P(∅) ≤ P(X ≤ x) ≤ P (R) = 1. Now, by the Monotone

Convergence Theorem,
Z Z
lim F (n) = lim χ(−∞,n] dP(x) = dP(x) = 1.
n→∞ n→∞ R R

3
CHAPTER 2. SOME MATHEMATICAL PRELIMINARIES 4

Similarly, for G(n) := 1 − F (−n), we have

Z
lim G(n) = lim (1 − χ(−∞,−n] )dPX (x) = 1.
n→∞ n→∞ R

Moreover, F is increasing by monotonicity of P and finally, again by Monotone Con-

vergence,
Z Z
lim+ 1 − F (x + h) + F (x) = lim+ (1 − χ(x,x+h] ) dP(x) = dP(x) = 1
h→0 h→0 R R

and so lim+ F (x + h) = F (x), i.e. F is right-continuous.

h→0
(b) Compute the expectation
Z Z Z
E(g(X)) = g(x) dP(x) = g(x)χ(−∞,x] dP(x) = g(x) dF (x).
R R R

(c) Compute the density of Bt2

√ √
F (u) := P(Bt2 ≤ u) = P(− u ≤ Bt ≤ u)
Z
= 2 √ p(y)dy
[0, u]
Z √
p( u)
=2 √ du
[0,u] 2 u
Z √
p( u)
= χ[0,∞) √ du.
(−∞,u] u
√
and so p(u) = χ[0,∞) p(√uu) where p(u) is the density of Bt .

3. Since Hi is a σ-algebra, ∅ ∈ Hi , ∀i ∈ I. So ∅ ∈ H = ∩i∈I Hi . If {Uj }j∈N ⊆ H, then

{Uj }j∈N ⊆ Hi for each i ∈ I and so Ω \ Uj ∈ Hi and ∪j∈A Uj ∈ Hi , ∀i ∈ I. Conclude that
Ω \ Uj ∈ H and ∪j∈A Uj ∈ H and H = ∩i∈I Hi is also a σ-algebra.

4. Let X : Ω 7→ R be a random variable with E(|X|p ) < ∞.

(a) Let A = {ω ∈ Ω | |X| ≥ λ > 0} and compute

Z Z Z
p p p p
E(|X| ) = |X| dP ≥ |X| dP ≥ λ dP = λp P(|X| ≥ λ).
Ω A A

(b) By Chebychev, P(|X| ≥ λ) = P(e|X| ≥ eλ ) ≤ 1

ekλ
E(ek|X| ) = M e−kλ .
CHAPTER 2. SOME MATHEMATICAL PRELIMINARIES 5

5. Since the measures are σ-finite, f (x, y) = xy is PX ⊗ PY measurable and E(|XY |) < ∞,
apply Fubini-Tonelli and compute
Z
E(XY ) = xy dPXY (x, y)
R2
Z
= xy dPX (x) ⊗ dPY (y)
R2
Z Z 
= y x dPX (x) dPY (y)
R R
Z
= E(X) y dPY (y)
R
= E(X)E(Y ).
P∞
6. (Borel-Cantelli) Let {Ak }∞
k=1 ⊆ F and suppose k=1 P(Ak ) < ∞. Then

P(∩∞ ∞
m=1 ∪k=m Ak ) ≤ lim sup P(Ak ) = 0
m→∞ k≥m

by dominated convergence.

7. Let Ω = ni=1 Gi .
F

(a) Note ∅ ∈ G and G is closed under unions by construction. It is also closed under
complements as Ω \ Gi = ∪j6=i Gj ∈ G.
(b) Write a new sequence defined by Fi = Gi \ ∪j≤i Fj and {Fi } will satisfy (a).
(c) Note that {X −1 (x ∈ R)} ⊆ F is disjoint. So, by (a) and (b), F is finite if and only if
all but finitely many X −1 (x ∈ R) are empty.

8. Let Bt be a 1-dimensional Wiener process.

(a) By Equation 2.2.3, since Bt ∼ N (0, t),

 2 
u u2
E(e iuBt
) = exp − V(Bt ) + iuE(Bt ) = e− 2 .
2

(b) Comparing power series coefficients, we deduce that

 2 n
(iu)2n 1 ut
E(Bt2n ) = − ,
(2n)! n! 2
(2n)! n
and so E(Bt2n ) = 2n n!
t .
CHAPTER 2. SOME MATHEMATICAL PRELIMINARIES 6

(c) Integrating by parts, compute the nth moment of Bt

Z
1 x2
2k
E(Bt ) = √ x2k e− 2t dx
2πt R
r Z √x x=∞ Z r Z √x
2t 2t
−u2 2t 2t 2
ue−u du
2k−1
 2k−2
=x ue du − (2k − 1)x
π x=−∞ R π
r Z  
2t −1 − x2
= −(2k − 1) x2k−2 e 2t dx
π R 2
Z
1 x2
= (2k − 1)t · √ x2k−2 e− 2t dx
2πt R
= (2k − 1)tE(Bt2k−2 ).
k−1 k
As E(Bt2 ) = t, we have that E(Bt2k ) = (2k)!t
2k k!
· t = (2k)!t
2k k!
.
2 2!·t
(d) Check the base case, n = 2k = 2, where E(Bt ) )] = 2·1! = t. If the claim is true for
n = 2k, then
(2k)!tk (2k + 2)!tk+1
E(Bt2k+2 ) = (2k − 1)tE(Bt2k ) = (2k + 1)t · = ,
2k k! 2k+1 (k + 1)!
and so it is also true for n = 2(k + 1) = 2k + 2, thus completing the induction step.
9. Note that {Xt } and {Yt } have the same distributions since neither distribution has any atoms
and they agree except on a zero set ∀t ≥ 0. Yet t 7→ Xt is discontinuous while t 7→ Yt is
continuous.
10. As Bt is Brownian, Bt+h − Bt ∼ N (0, h). Since h is fixed, {Bt+h − Bt }h≥0 have the same
distributions ∀t ≥ 0.
 
(1) (2) (n) (j)
11. As B0 = B0 , B0 , . . . B0 = 0, B0 = 0 for all j ∈ {1, . . . n}. Bt is almost surely
continuous only if its components are almost surely continuous. Each component is normally
(i) (j)
distributed with E(Btj ) = 0 as E(Bt ) = ~0 and Cov(Bt , Bt ) = tδij as Cov(Bt ) = tI.
12. Let Wt := Bs+t − Bs where s ≥ 0 is fixed. Then W0 = Bs − Bs = 0 and Wt is almost
surely continuous as the sum of two almost surely continuous stochastic processes. Noting
Wt2 − Wt1 = Bs+t2 − Bs+t1 is independent of both Bs+t1 and Bs , deduce that Wt2 − Wt1 is
independent of Wt1 = Bs+t1 − Bs . The expected value is E(Wt ) = E(Bs+t ) − E(Bs ) = 0
and the variance is
V(Wt ) = E((Bs+t − Bs )2 )
2
= E(Bs+t ) − 2E(Bs Bs+t ) + E(Bs2 )
2
= E(Bs+t ) − 2E(Bs (Bs+t − Bs )) − E(Bs2 )
2
= E(Bs+t ) − 2E(Bs )E(Bs+t − Bs ) − E(Bs2 )
= (s + t) − 0 − s
= t.
CHAPTER 2. SOME MATHEMATICAL PRELIMINARIES 7

Since Wt is the sum of two normal distributions, it is also normal and Wt ∼ N (0, t).

13. Compute
ρ2
Z Z ρ Z
1 − |~x|2 2 2π 2
− r2t
2t ρ2
P0 (Bt ∈ Dρ ) = e 2t d ~x = re dr = e−u du = 1 − e− 2t .
|x|<ρ 2πt 2πt 0 0

14. Compute
Z  Z
Ex χK (Bt ) dt = P(Bt ∈ K) dt
[0,∞] [0,∞]
Z Z 
1 − 2t
|~
x−~y |2
n
= n/2
e d ~x dt
[0,∞] K (2πt)
Z
1 |~
x−~y |2
− 2t
≤
n/2
e µ(K) dt
[0,∞] (2πt)

∞
=0

and deduce that the expected total time spent in K is 0.

15. Note that U U T = I, whence | det U | = 1 and the probability measures are identical by
change of variables. It follows that both are Brownian.

16. Let Wt = 1c Bc2 t . We have W0 = B0 = 0 and that Wt is absolutely continuous as a scaling

of absolutely continuous Bt . Finally,

P0 (Wt ∈ U ) = P0 (Bc2 t ∈ cU )
Z
= p(c2 t, 0, y) dy
ZcU
1
= p(t, 0, y/c) dy
cU c
Z
1
= p(t, 0, y 0 )(cdy 0 )
U c
= P0 (Bt ∈ U ),

and so Wt is also a Brownian motion.

17. Let Xt (·) be a continuous stochastic process.

CHAPTER 2. SOME MATHEMATICAL PRELIMINARIES 8

(a) Recall that E(Bt ) = 0, E(Bt2 ) = t and E(Bt4 ) = 3t2 . Then

! !
X
2

2 X
2

2
E ( ∆Bk − ∆tk ) = E ( ∆Bk − ∆tk )
k k
X
E(∆Bk4 ) − 2∆tk E(∆Bk2 ) + ∆t2k


=
k
X
3∆t2k − 2∆t2k + ∆t2k


=
k
X
=2 ∆t2k .
k

(2)
So hB, Bit (w) = t.
(b) Note that the Brownian motion has positive quadratic variation t on [0, t]. So
(2)
(1) hB, Bit (w)
hB, Bit (w) ≥ lim = ∞.
k∆Bk k→0+ k∆Bk k
Chapter 3

Itô Integrals

1. Compute
dnte
Z t t
−1
X jt
s dBs = lim (B (j+1)t − B jt )
0 n→∞
j=0
n n n

dnte
t
−1
dnte t X t
= lim B dnte − lim B jt + lim (B0 − B dnte )
n→∞ n n n→∞ n n n→∞ n n
j=0
Z t
= tBt − Bs ds.
0

2. Compute
dnte
Z t t
−1
X
Bs2 dBs = lim B 2jt (B (j+1)t − B jt )
0 n→∞ n n n
j=0
dnte
t
−1
X 1 1 3 1

3 2 3
= lim B (j+1)t − B j − B jt (B (j+1)t − B j ) − (B (j+1)t − B j )
n→∞
j=0
3 n 3 n n n n 3 n n

 dnte 
t
−1
1 X t
= Bt3 − lim  B jt + O(t2 /n)
3 n→∞
j=0
n n

Z t
1 3
= Bt − Bs ds.
3 0

(X)
3. Let {Nt } be some filtration and let {Ht } be the filtration of process Xt .

(a) Compute
E(Xt | Hs(X) ) = E E(Xt | Ns ) | Hs(X) = E(Hs | Hs(X) ) = Hs .

9
CHAPTER 3. ITÔ INTEGRALS 10

(b) Compute
(X)
E(Xt ) = E(E(Xt | H0 )) = E(X0 ).

(c) Let Y ∼ Bernoulli(0.5) and fix X0 = 2Y − 1. Then Xt = t · sgn(X0 ) satisfies

E(Xt ) = E(X0 ) = 0, but E(Xt | Fs ) = t · sgn(X0 ) 6= s · sgn(X0 ).

4. Compute

E(Bt + 4t | Fs ) = Bs + 4t 6= Bs + 4s
E(Bt2 | Fs ) = E((Bt − Bs )2 + 2Bs (Bt − Bs ) + Bs2 | Fs ) = Bs2 + t − s 6= Bs2
 Z t  Z s Z t Z s
2 2 2
E t Bt − 2 uBu du | Fs = t Bs − 2 uBu du − 2 uBs du = s Bs − 2 uBu du
0 0 s 0
(1) (2) (1) (2)
E(Bt Bt | Fs ) = E(Bt | Fs )E(Bt | Fs ) = Bs(1) Bs(2) ,

and deduce that only the last two are martingales.

5. Verify E(|Bt2 − t|) ≤ E(Bt2 ) + t = 2t < ∞ and compute

E(Bt2 − t | Fs ) = E((Bt − Bs )2 + 2Bs (Bt − Bs ) + Bs2 − t | Fs ) = Bs2 + t − s − t = Bs2 − s.

to deduce that Xt := Bt2 − t is a martingale.

p p √
6. Verify E(|Bt3 − 3tBt |) ≤ E(Bt2 )( E(Bt4 ) + 3t) = (3 + 3)t3/2 < ∞ and compute

E(Bt3 − 3tBt | Fs ) = E((Bt − Bs )3 + 3Bs (Bt − Bs )2 + 3Bs2 (Bt − Bs ) + Bs3 − 3tBs | Fs )

= 3Bs (t − s) + Bs3 − 3tBs
= Bs3 − 3sBs

to deduce that Yt := Bt3 − 3tBt is a martingale.

7. In this question, the formula for Itô iterated integrals is derived.

R that {0 ≤ u1 · · · ≤ un } is Borel
(a) Note  measurable and χ0≤u1 ···≤un is Ft -adapted. Finally
T
E 0 f (t1 , . . . tn , ω) dt1 . . . dtn ≤ T n < ∞.
2

(b) For n ∈ {1, 2, 3}

Z t  
1/2 Bt
1! dBu = Bt = t H1 √
0 t
Z tZ v Z t  
2 B2
2! dBu dBv = 2 Bv dBv = Bt − t = tH2 √
0 0 0 t
Z tZ wZ v Z t  
2 3 3/2 Bt
3! dBu dBv dBw = 3 (Bw − w) dBw = Bt − 3tBt = t H3 √ .
0 0 0 0 t
CHAPTER 3. ITÔ INTEGRALS 11

(c) Deduce that d(Bt3 − 3tBt ) = 3(Bt2 − t) dBt and so Yt := Bt3 − 3tBt is a martingale.

8. There exists continuous martingale Mt iff there exists Y ∈ L1 such that Mt = E(Y | Ft ).
(a) Verify that E(|E(Y | Ft )|) ≤ E(E(|Y | | Ft ) = E(|Y |) < ∞ and
E(Mt | Fs ) = E(E(Y | Ft ) | Fs ) = E(Y | Fs ) = Ms .

(b) If Mt is a continuous martingale such that supt>0 E(|X|p ) < ∞ for p ∈ (1, ∞), then
∃M such that kMt − M kL1 → 0 as t → ∞. So let Y = M and
Z Z
lim |Ms − E(M | Fs )| dP = lim |E(Ms − M | Fs )| dP
s→∞ Ω s→∞ Ω
s s
Z
≤ lim E(|Ms − M | | Fs ) dP
s→∞ Ω
s
Z
= lim |Ms − M | dP
s→∞ Ωs
= 0.

9. Compute
bntc
Z T t
−1
X 1
Bt ◦ dBt = lim (B jt + B (j+1)t )(B (j+1)t − B jt )
0 n→∞
j=0
2 n n n n

bntc bntc
t
−1 t
−1
X X 1
= lim B jt (B (j+1)t − B jt ) + lim (B (j+1)t − B jt )2
n→∞
j=0
n n n n→∞
j=0
2 n n

1 t t
= Bt2 − +
2 2 2
1 2
= Bt .
2
10. If f (t, ω) varies smoothly in t, then the Itô and Stratonovich integrals coincide. Compute
Z T Z T
1
f (t, ω) ◦ dBt = f (t, ω) dBt + hf (t, ω), Bt i(2)
0 0 2
and
E(hf (t, ω), Bt i(2) )2 ≤ E(hBt , Bt i(2) E(hf (t, ω), f (t, ω)i(2)
T
≤ T lim + sup (K|∆tk |1+ε )
k∆tk k→0 |∆tk | |∆tk |

= KT 2 lim k∆tk kε
k∆tk k→0+

= 0.
CHAPTER 3. ITÔ INTEGRALS 12

(N )
11. Define white noise Wt = max{−N, min{Wt , N }}. Since Wt and Ws are independent
(N ) (N )
and identically distributed, it follows that Wt and Ws are as well. If Wt is continuous,
(N )
then since |Wt | ≤ N and by bounded convergence
(N ) 2 (N )
lim 2E(Wt ) = lim E(|Wt − Ws(N ) |2 ) = 0.
t→s t→s

a.s.
But then Wt = 0, which is a contradiction.

12. Let ◦dBt denote the Stratonovich differential.

α2
(i) Since αXt ◦ dBt = 2
Xt dt + αXt dBt ,

α2
dXt = (γ + )Xt dt + αXt dBt .
2
Since (t2 + cos(Xt )) ◦ dBt = − sin(X
2
t)
(t2 + cos(Xt )) dt + (t2 + cos(Xt ) dBt ,

sin(Xt )
dXt = (cos(Xt ) − t2 ) dt + (t2 + cos(Xt )) dBt .
2
α2
(ii) Since αXt dBt = αXt ◦ dBt − 2
Xt dt,

α2
dXt = (r − )Xt dt + αXt ◦ dBt .
2
Since Xt2 dBt = Xt2 ◦ dBt − Xt3 dt,

dXt = (2e−Xt − Xt3 ) dt + Xt2 ◦ dBt .

13. Let Xt be continuous in mean square. Calculate

(a) lim E[(Bt − Bs )2 ] = lim E[(Bt−s )2 ] = lim(t − s) = 0

s→t s→t s→t

(b) lim E[(f (Bt ) − f (Bs )) ] ≤ lim C E[(Bt − Bs )2 ] = 0

2 2
s→t s→t
(c) and finally by Itô isometry,
"Z
T 2 # Z T 
2
lim E (Xs − φn (s)) dBs = lim E (Xs − φn (s)) ds
n→∞ S n→∞ S
(n+1)
" #
XZ tj
= lim E (Xt − Xt(n) )2 dt
n→∞ (n)
tj j
j

≤ (T − S) lim sup E[(Xt − Xt(n) )2 ]

n→∞ 1≤j≤n j

= 0.
CHAPTER 3. ITÔ INTEGRALS 13

14. Show that h(ω) is Ft measurable if and only if it is the pointwise limit of a sum-product of
bounded continuous functions g(Btj ).

(a) Assume that h is bounded since {hn (ω) := h(ω)1{|h(ω)|<n} } converges pointwise to h.
(b) Let Hn be the σ-algebra generated by B(tj ) for tj = 2jn ≤ t. Then Ft = σ (∪n Hn )
and so by Corollary (C.9), h = E[h|Fn ] = lim E[h|Hn ].
n→∞


(c) By Doob-Dynkin, E[h|Hn ](ω) = g Bt1 , . . . B(tb2n tc ) . Since C(Rk ) is dense in
L1 (Rk ) and by Stone-Weierstrass P (Rk ) is dense in C(Rk ), a limiting sequence must
exist.
RT RT
15. Suppose C + S f (t, ω) dBt (ω) = D + S g(t, ω) dBt (ω). Then we have that
Z T Z T 
C − D = E[C − D] = E g(t, ω) dBt (ω) − f (t, ω) dBt (ω) = 0 =⇒ C = D,
S S

and by Itô isometry,

"Z
T Z T 2 # Z T
0=E g(t, ω) dBt (ω) − f (t, ω) dBt (ω) = E[(g(t, ω) − f (t, ω))2 ] ds,
S S S

whence g(t, ω) = f (t, ω) almost surely for (t, ω) ∈ [S, T ] × Ω.

16. By Jensen’s inequality, E [E[X|H]2 ] ≤ E [E[X 2 |H]] = E[X 2 ].
17. Let G be a finite σ-algebra with partition Ω = ni=1 Gi .
F

(a) Note that E[X|G](ω) = ni=1 ci 1Gi (ω) = ci on Gi .

P

(b) Show that

R ! R
X dP X dP
Z Z Z
Gi Gi
dP = 1 dP = X dP, ∀i ∈ {1, . . . n}.
Gi P(Gi ) P(Gi ) Gi Gi
R
Gi X dP
(c) By part (b), ci = P(Gi )
. Show for ω ∈ Gi that
n
R
X dP
1Gi (ω)
X Gi
E[X|G](ω) =
i=1
P(Gi )
R
X dP
= Gi
P(G )
Pm i
ak P(X = ak , ω ∈ Gi )
= k=1
P(Gi )
Xm
= ak P(X = ak |Gi ).
k=1
Chapter 4

The Itô Formula

1. Compute

(a) dXt = d(Bt2 ) = 2Bt dBt + d[B, B]t = 2Bt dBt + dt

(b) dXt = d(2 + t + eBt ) = (1 + 21 eBt ) dt + eBt dBt
 
(1) (2) (1) (1) (2) (2)
(c) dXt = d (Bt )2 + (Bt )2 = 2Bt dBt + 2Bt dBt + 2 dt
(d) dXt = d((t0 + t, Bt )) = (dt, dBt )
(e) and finally
(1) (2) (3) (2) (1) (3)
dXt = d((Bt + Bt + Bt , (Bt )2 − Bt Bt ))
(1) (2) (3) (2) (2) (3) (1) (1) (3)
= (dBt + dBt + dBt , 2Bt dBt + dt − Bt dBt − Bt dBt ).

2. Using Itô’s Lemma, differentiate

 Z t 
1 3
d B − Bs ds = Bt2 dBt + Bt d[B, B]t − Bt dt = Bt2 dBt
3 t 0

and deduce that

Z t Z t
1
Bs2 dBs = Bt3 − Bs ds.
0 3 0

3. Let Xt and Yt be Itô processes. Then, letting f (t, x, y) = xy and by Itô’s formula

d(Xt Yt ) = ft (t, Xt , Yt ) dt + fx (t, Xt , Yt ) dXt + fy (t, Xt , Yt ) dYt

1 1
+ fxx (t, Xt , Yt ) d[X, X]t + fxy (t, Xt , Yt ) d[X, Y ]t + fyy (t, Xt , Yt ) d[Y, Y ]t
2 2
= Yt dXt + Xt dYt + d[X, Y ]t

14
CHAPTER 4. THE ITÔ FORMULA 15

and deduce the integration of parts formula

Z t Z t
Xs dYs = (d(Xs Ys ) − Ys dXs − d[X, Y ]s )
0 0
Z t Z t
= Xt Yt − X0 Y0 − Ys dXs − d[X, Y ]s .
0 0
R 
t 1 2
4. Let Zt = exp 0
hθ(s, ω), dBs i − 2 |θ(s, ω)| ds .

(a) Then, letting Zt = eYt and by Itô’s formula,

1
dZt = eYt dYt + eYt d[Y, Y ]t
2 !
n
1 1 X
= Zt hθ(t, ω), dBt i − |θ(t, ω)|2 dt + θi (s, ω) dB (i) , θj (s, ω) dB (j) s
 
2 2 i,j=1
= Zt hθ(t, ω), dBt i.

(b) It suffices to check that

 Z t 2
2
 
[E(|Zt |)] = E  dZs 

0
 Z t 2
 
= E  Zs hθ(s, ω), dBs i

0
Z tX n
!2
≤E |Zs θi (s, ω)| dBs(i)
0 i=1
n
!
X Z t
=E |Zs θi (s, ω)||Zs θj (s, ω)| d[B (i) , B (j) ]s
i,j=1 0
n
X Z t 
2
= E |Zs θi (s, ω)| ds
i=1 0

< ∞.

5. Let βk (t) = E(Btk ). Then, by Itô’s lemma,

1
dBtk = kBtk−1 dBt + k(k − 1)Btk−2 dt
2
and so
Z t  Z t  Z t 
11
βk (t) = E(Btk )=E dBsk
= E k(k − 1)Btk−2
ds = k(k − 1) βk−2 (s) ds.
0 0 22 0
Rt 2 Rt
Deduce that β4 (t) = 6 0 β2 (s) ds = 6 · t2 = 3t2 and β6 (t) = 15 0 3s2 ds = 15t3 .
CHAPTER 4. THE ITÔ FORMULA 16
Pn (j)
6. Define geometric Brownian motions Xt = ect+αBt and Yt = ect+ j=1 αj Bt
.

(a) Calculate
1
dXt = cect+αBt dt + αect+αBt dBt + α2 ect+αBt d[B, B]t
 2
α2

= Xt (c + ) dt + α dBt .
2
(b) Calculate
n n
!
X (j) 1X
dYt = Yt c dt + αj dBt + αi αj d[B (i) , B (j) ]t
j=1
2 i,j=1
n n
!
1X 2 X (j)
= Yt (c + α ) dt + αj dBt .
2 j=1 i j=1

7. Let Xt solve dXt = v(t, ω) dBt .

(a) Note that Bt is a martingale while Bt2 is not.

Rt
(b) Define Mt = Xt2 − 0 v(s, ω)2 ds. Then
dMt = 2Xt dXt + [dX, dX]t − v(t, ω)2 , dt
= 2Xt v(t, ω) dBt + v(t, ω)2 − v(t, ω)2 dt

= 2Xt v(t, ω) dBt .

Moreover,
Z t 
E(|Mt |) ≤ E(Xt2 ) +E 2
v(s, ω) ds
0
Z t 2 Z t 
2
=E v(s, ω) dBs +E v(s, ω) ds
0 0
Z t 
2
= 2E v(s, ω) ds
0
< ∞.

8. Let f (x(1) , . . . x(n) ) be a function of class C 2 .

(a) By Itô’s lemma,
n n
X (i) 1X 2
d(f (Bt )) = ∂i f (Bt ) dBt + ∂ f (Bt ) d[B (i) , B (j) ]t
i=1
2 i,j=1 ij
1
= h∇f (Bt ), dBt i + ∆f (Bt ) dt
2
CHAPTER 4. THE ITÔ FORMULA 17

and so
Z t Z t Z t
1
f (Bt ) − f (B0 ) = d(f (Bs )) = h∇f (Bs ), dBs i + ∆f (Bs ) ds.
0 0 2 0

(b) Assume that g is of class C 1 everywhere, as well as C 2 and uniformly bounded outside
of finitely many points with |g 00 (z)| ≤ M for z ∈/ {z1 , . . . zk }. Then the set of functions
{f } of class C 2 uniformly bounded with |f 00 (z)| ≤ M are C k -dense. So we can extract
a sequence {fk } such that fk ⇒ g, fk0 ⇒ g 0 as well as fk00 → g 00 and |fk00 | ≤ M on
R \ {z1 , . . . zk }. So
Z t
1 t 00
 Z 
0 0 00
 
lim (fk − g)(Bt ) + (fk − g)(0) +
 (fk − g ) dBs + (fk − g ) ds
k→∞ 0 2 0
1 t 00
Z
0 0
≤ lim |(fk − g)(Bt )| + |(fk − g)(0)| + tkfk − g k∞ + |f − g 00 | ds
k→∞ 2 0 k
= 0,

where the last term vanishes by bounded convergence.

9. Clearly
Z t Z t∧τn
∂gn ∂g
v (s, Xs )χs≤τn dBs = v (s, Xs ) dBs
0 ∂x 0 ∂x

and the result follows by Itô’s lemma where dXt = u dt + v dBt . Since E(|Xt |) < ∞, it
follows that lim P(τn > t) = lim P(Xt < n) = 1 and so the identity holds almost surely.
n→∞ n→∞

10. (Tanaka) In this problem, Tanaka’s formula for Brownian motion is derived.

(a) Substitute u ≡ 0 and v ≡ 1 here. Then as gε00 (x) = 1ε χ|x|<ε (x)

t t
d2 gε
Z Z
1 1 1
2
(Bs ) ds = χ|Bs |<ε ds = |{s ∈ [0, t] | |Bs | < ε}|.
2 0 dx 2ε 0 2ε

(b) Differentiate to get

Z t Z t
Bs
gε0 (Bs )χ|Bs |<ε dBs = χ|Bs |<ε dBs ,
0 0 ε
and apply Itô isometry to get
t 2 t t
Bs2
Z Z  Z
Bs
lim+ E χ|Bs |<ε dBs = lim+ E χ|Bs |<ε ds ≤ lim+ P(|Bs | < ε) ds = 0.
ε→0 0 ε ε→0 0 ε2 ε→0 0
CHAPTER 4. THE ITÔ FORMULA 18

(c) As ε → 0 for g(x) = x,

Z t
1
|Bt | = |B0 | + lim+ sgn(Bs )χ|Bs |≥ε ds + lim+ |{s ∈ [0, t] | |Bs | < ε}|
ε→0 ε→0 2ε
Z t 0
= |B0 | + sgn(Bs ) ds + Lt .
0

11. Let Xt = e t/2

cos(Bt ), Yt = et/2 sin(Bt ) and Zt = (Bt + t)e−Bt −t/2 . Compute
(a) dXt = 21 et/2 cos(Bt ) dt−et/2 sin(Bt ) dBt + 12 (−et/2 cos(Bt )) d[B, B]t = −et/2 sin(Bt ) dBt
(b) dYt = 21 et/2 sin(Bt ) dt+et/2 cos(Bt ) dBt + 12 (−et/2 sin(Bt )) d[B, B]t = et/2 cos(Bt ) dBt
(c) and finally
dZt = e−Bt −t/2 d(Bt + t) + (Bt + t)d(e−Bt −t/2 ) + d[Bt + t, e−Bt −t/2 ]
1 1
= e−Bt −t/2 (dt + dBt ) − Xt dt − Xt dBt − e−Bt −t/2 dt + (Bt + t)e−Bt −t/2 dt
2 2
= e−Bt −t/2 (1 − t − Bt ) dBt .
12. The given condition implies E(|Xt |) < ∞. So Xt is a martingale if and only if E(Xt | Fs ) =
Xs . Then
Z t
E( u(r, ω) dr | Fs ) = E(Xt − Xs | Fs ) = 0.
s

Moreover by dominated convergence

Z t
d
E(u(t, ω) dr | Fs ) = E( u(r, ω) dr | Fs ) = 0.
ds s

Then
u(t, ω) = E(u(t, ω) | Ft ) = lim− E(u(t, ω) | Fs ) = 0.
s→t

13. Let dXt = u(t, ω) dt + dBt where u(t, ω) ∈ V([0, T ]). Then Yt = Xt Mt is a martingale,
where
 Z t
1 t 2
Z 
Mt = exp − u(r, ω) dBr − u (r, ω) dr
0 2 0
√ qR t
 
since E(|Mt |) < ∞ (see question 4b), E(|Xt |) ≤ t 0
u2 (r, ω) dr + 1 < ∞ and

d(Xt Mt ) = Mt dXt + Xt dMt + d[X, M ]t

1
= Mt (u(t, ω) dt + dBt ) + Mt Xt (−u(t, ω) dBt − u2 (t, ω) dt)
2
1
− Mt u(t, ω) dt + Mt Xt u2 (t, ω) dt
2
= Mt (1 − u(t, ω)Xt ) dBt .
CHAPTER 4. THE ITÔ FORMULA 19

14. In this problem, the martingale representation of stochastic processes is explicitly shown.

(a) Compute dFt = dBt , E(FT ) = 0 and

dFt − dE(Ft ) = 1 dBt =⇒ f (t, ω) = 1.

(b) Compute dFt = Bt dt, E(FT ) = 0 and

dFt − dE(Ft ) = Bt dt = d(T BT ) − t dBt = (T − t) dBt =⇒ f (t, ω) = T − t.

(c) Compute dFt = 2Bt dBt + dt, E(FT ) = T and

dFt − dE(Ft ) = 2Bt dBt + 1 dt − 1 dt = 2Bt dBt =⇒ f (t, ω) = 2Bt .

(d) Compute dFt = 3Bt2 dBt + 3Bt dt, E(FT ) = 0 and

dFt − dE(Ft ) = 3Bt2 dBt + 3Bt dt

= 3Bt2 + 3(T − t)) dBs =⇒ f (t, ω) = 3Bt2 + 3T − 3t.

(e) Recall that eBt −t/2 is a martingale and compute

d(eBt −t/2 ) = eBt −t/2 dBt .

Deduce that
 Z T 
Bt −t/2
e BT
=e T /2
1+ e dBt =⇒ f (t, ω) = eBt +(T −t)/2 .
0

(f) Find martingale et/2 sin(Bt ) and compute

d(et/2 sin(Bt )) = et/2 cos(Bt ) dBt

Deduce that
Z T
−T /2
sin(BT ) = e et/2 cos(Bt ) dBt =⇒ f (t, ω) = e−(T −t)/2 cos(Bt ).
0

15. Define Xt = (x1/3 + 13 Bt )3 . Then

 
1
2/3 1/3 1 1
dXt = 3Xt d(x1/3 1/3 1/3
+ Bt ) + 3Xt d x + Bt , x + Bt
3 3 3
2/3 1 1/3
= Xt dBt + Xt dt.
3
Chapter 5

Stochastic Differential Equations

1. Compute
B
(a) dXt = d(eBt ) = eBt dBt + 12 t d[B, B]t = 12 Xt dt + Xt dBt
Bt 1 Bt 1 1


(b) dXt = d 1+t = 1+t dBt − (1+t) 2 dt = 1+t dBt − 1+t Xt dt

(c) dXt = d(sin(Bt )) = cos(Bt ) dBt − 12 sin(Bt ) dt = cos(Bt ) dBt − 12 Xt dt

(1)
(d) dXt = dt and
(2) (2)
dXt = d(et Bt ) = et dBt + et Bt dt = et dBt + Xt dt.

(e) and finally differentials

1
d(cosh(Bt )) = sinh(Bt ) dBt + cosh(Bt ) dt
2
and
1
d(sinh(Bt )) = cosh(Bt ) dBt + sinh(Bt ) dt
2
to deduce that
! ! !
(1) (1) (2)
dXt 1 Xt Xt
(2) = (2) dt + (1) dBt .
dXt 2 Xt Xt

(1) (2)
2. Let Xt = a cos(Bt ) and Xt = b sin(Bt ). Then

(1) a 1 (1) a (2)

dXt = −a sin(Bt ) dBt − cos(Bt ) dt = − Xt dt − Xt dBt
2 2 b
and
(2) b 1 (2) b (1)
dXt = b cos(Bt ) dBt − sin(Bt ) dt = − Xt dt + Xt dBt .
2 2 a
20
CHAPTER 5. STOCHASTIC DIFFERENTIAL EQUATIONS 21

3. The solution is given by

n n
!
1X 2 X
Xt = X0 exp (r − α )t + αk dBk .
2 k=1 k k=1

4. In this problem, solutions to stochastic differential equations are found.

(1) (1) (1) (1) (1)
(a) The solution to dXt = dt + dBt is Xt = X0 + t + Bt and
(2) (1) (2) (1) (1) (2)
dXt = Xt dBt = (X0 + t + Bt ) dBt

is
Z t
(2) (2) (1) (2)
Xt = X0 + X0 Bt + (s + Bs(1) ) dBs(2) .
0

(b) Using integrating factors, solve dXt = Xt dt + dBt for

Z t
−t
e Xt − X0 = e−s dBs
0

and deduce that the solution Xt is

Z t
t
X t = e X0 + et−s dBs .
0

(c) Using integrating factors, solve dXt = −Xt dt + e−t dBt for
Z t
t
e Xt − X0 = dBs
0

and deduce that the solution Xt is

Xt = e−t (X0 + Bt ).

5. The Langevin equation is given by

dXt − µXt dt = σdBt .

(a) Using integrating factors, solve for

Z t
−µt
e Xt − X0 = e−µs σ dBs
0

and deduce that the solution Xt is

Z t
µt
Xt = e X0 + σ eµ(t−s) dBs .
0
CHAPTER 5. STOCHASTIC DIFFERENTIAL EQUATIONS 22

(b) The expected value of Xt is

E(Xt ) = eµt X0
and, by Itô isometry, the variance of Xt is
Z t 2 !  Z t
σ 2 2µt

2 µ(t−s) 2 2µ(t−s)
V(Xt ) = E σ e dBs =E σ e ds = (e − 1).
0 0 2µ

6. Suppose Yt is given by
dYt = r dt + αYt dBt .
Using integrating factors, solve for
α2
 
−αBt −αBt
d(e Yt ) = e Yt r− dt
2
and
2
Z t
α2
−αBt + α2 t
e Yt − Y0 = re−αBs + 2
s
ds.
0

Deduce that
2
Z t
α2
αBt − α2 t
Yt = e Y0 + r eα(Bt −Bs )− 2
(t−s)
ds.
0

7. The Ornstein-Uhlenbeck process is given by

dXt = (m − Xt ) dt + σ dBt .

(a) Using integrating factors, solve for

Z t Z t
t s
e Xt − X0 = e m ds + es σ dBs
0 0

and deduce that the solution Xt is

Z t
−t −t
Xt = e X0 + m(1 − e ) + σ es−t dBs .
0

(b) The expected value of Xt is

E(Xt ) = m + e−t (X0 − m)
and the variance of Xt is
Z t 2 !  Z t
σ2

V(Xt ) = E σ 2 s−t
e dBs =E σ 2
e 2s−2t
ds = (1 − e−2t ).
0 0 2
CHAPTER 5. STOCHASTIC DIFFERENTIAL EQUATIONS 23

8. Consider the stochastic differential equation

!  ! !
(1)  (1) (1)
dXt 0 1 Xt α dBt
(2) = (2) dt + (2) .
dXt −1 0 Xt β dBt
By d’Alembert’s formula, it has a solution of the form
Z t
At
Xt = e X0 + eA(t−s) g(s) ds,
0
where
   it  −1  
At 1 1 e 0 1 1 cos(t) sin(t)
e = = .
i −1 0 e−it i −1 − sin(t) cos(t)
Conclude that the solutions are
Z t Z t
(1) (1) (2)
Xt = X0 cos(t) + X0 sin(t) + α cos(t − s) dBs(1) +β sin(t − s) dBs(2)
0 0
and
Z t Z t
(2) (1) (2)
Xt = −X0 sin(t) + X0 cos(t) − α sin(t − s) dBs(1) +β cos(t − s) dBs(2) .
0 0

9. Let dXt = ln(1 + Xt2 ) dt + χ{Xt >0} Xt dBt . It suffices to check that
2
|b(t, x)| + |σ(t, x)| = ln(1 + x2 ) + χ{x>0} |x| ≤ (|x| + 1) + |x| ≤ 2(|x| + 1),
e
2 2
E(|X0 | ) = α < ∞, and
|b(t, x) − b(t, y)| + |σ(t, x) − σ(t, y)| ≤ | ln(x2 ) − ln(y 2 )| + |x − y| ≤ 3|x − y|.
Hence, by Theorem 5.2.1, there is a unique strong solution to the stochastic differential
equation.
10. Calculate
 Z t Z t 2
E(Xt2 ) =E Z+ b(s, Xs ) ds + σ(s, Xs ) dBs
0 0
Z t 2 Z t 2 !
≤ 3 E(Z 2 ) + E b(s, Xs ) ds +E σ(s, Xs ) dBs
0 0
 Z t  Z t 
2 2 2
≤ 3 E(Z ) + T E b(s, Xs ) ds + E σ(s, Xs ) ds
0 0
 Z t 
2 2 2
≤ 3E(Z ) + 6C T + E(|Xs | ) ds (T + 1)
0
Z t
2 2 2
= (3E(Z ) + 6C T (T + 1)) + 6C (T + 1) E(|Xs |2 ) ds.
0
and apply Gronwall to derive the result.
CHAPTER 5. STOCHASTIC DIFFERENTIAL EQUATIONS 24

11. Consider the stochastic process

Z t
dBs
Yt = a(1 − t) + bt + (1 − t) .
0 1−s

Then Y0 = a and, for t ∈ [0, 1), Yt solves

Z t
dBs dBt
dYt = (b − a) dt − dt + (1 − t)
0 1−s 1−t
 Z t 
1 dBs
= (b − a)(1 − t) − (1 − t) dt + dBt
1−t 0 1−s
 Z t 
1 dBs
= b − a(1 − t) − bt − (1 − t) dt + dBt
1−t 0 1−s
b − Yt
= dt + dBt .
1−t
 R t dBs 2 Rt 1
Finally by Itô isometry E (1 − t)2 0 1−s = (1 − t)2 0 (1−s)2 ds = (1 − t)t → 0 as
a.s.
t → 1− and so limit lim− Yt = b.
t→1

12. Let y 00 (t) + (1 + εWt )y(t) = 0 where Wt = dBt

dt
is 1-dimensional white noise.

(a) Rewrite
       
dyt 0 1 yt 0 0 yt
= dt + dBt .
dy˙t −1 0 y˙t −ε 0 y˙t
Rt Rt
(b) Check that, if y(t) = y(0) + y 0 (0)t + 0
(r − t)y(r) dr + 0
ε(r − t)y(r) dBr , then
Z t Z t Z t
0 0 0
y (t) = y (0) − y(r) dr − εy(r) dBr = y (0) − y(r)(1 + εWr ) dr
0 0 0

and y 00 (t) = −(1 + εWr ) dr.

13. Let x00t + a0 x0t + w2 xt = (T0 − α0 x0t )ηWt where Wt is 1-dimensional white noise. Then
         
dxt 0 1 xt 0 0 xt 0
= 2 dt + dBt + dBt
dẋt −w −a0 ẋt 0 −α0 η ẋt T0 η

and by d’Alembert’s formula the solution is

Z t Z t
At A(t−s)
Xt = e X0 + e KXs dBs + eA(t−s) M dBs .
0 0
CHAPTER 5. STOCHASTIC DIFFERENTIAL EQUATIONS 25
q
a2
The eigenvalues of A satisfy λ2 +a0 λ+w2 = 0 and are λ± = − a20 ± w2 − 40 i =: −λ±ξi.
Then take the exponential of matrix A
  λ t  −1
At 1 1 e + 0 1 1
e =
λ+ λ− 0 e λ− t λ+ λ−
 
1 λ− eλ+ t − λ+ eλ− t eλ− t − eλ+ t
=
λ− − λ+ −λ− λ+ (eλ− t − eλ+ t ) λ− eλ− t − λ+ eλ+ t
1 e−λt (−λ · 2i sin(ξt) − ξi · 2 cos(ξt) e−λt (−2i sin(ξt))
 
=−
2ξi −w2 e−λt (−2i sin(ξt)) e−λt (−λ · 2i sin(ξt) − ξi · 2 cos(ξt) + 2λ · 2i sin(ξt))
e−λt λ sin(ξt) + ξ cos(ξt)
 
sin(ξt)
=
ξ −w2 sin(ξt) λ sin(ξt) + ξ cos(ξt) − 2λ sin(ξt)
e−λt
= ((λ sin(ξt) + ξ cos(ξt))I + A sin(ξt)) .
ξ
Next, letting ys = x˙s , gt = e−λt sin(ξt)
ξ
and ht = e−λt ξ cos(ξt)−λ
ξ
sin(ξt)
, compute

α0 ηe−λ(t−s) 0
    
A(t−s) sin(ξ(t − s)) xs −α0 ηys gt−s
e KXs = − =
ξ 0 ξ cos(ξ(t − s)) − λ sin(ξ(t − s)) x˙s −α0 ηys ht−s
and
T0 ηe−λ(t−s)
   
A(t−s) sin(ξ(t − s)) ηT0 gt−s
e M= = .
ξ ξ cos(ξ(t − s)) − λ sin(ξ(t − s)) ηT0 ht−s
It follows that
Z t
xt = η (T0 − α0 ys )gt−s dBs
0

and
Z t
yt = η (T0 − α0 ys )ht−s dBs .
0

(1) (2)
14. Letting Zt = F (Bt ), where Bt = Bt + iBt , calculate
(1) (2)
dZt = Fx (Bt ) dBt + Fy (Bt ) dBt
1
+ Fxx (Bt ) d[B (1) , B (1) ]t + Fxy (Bt ) d[B (1) , B (2) ]t + Fyy (Bt ) d[B (2) , B (2) ]t
2
(1) (2) 1
= (ux + ivx ) dBt + (uy + ivy ) dBt + (uxx + ivxx + uyy + ivyy ) dt
2
1
= hF 0 (Bt ), dBt i + (vxy − iuxy + uyy + ivyy ) dt
2
1
= hF 0 (Bt ), dBt i + (−uyy − ivyy + uyy + ivyy ) dt
2
= hF 0 (Bt ), dBt i.
CHAPTER 5. STOCHASTIC DIFFERENTIAL EQUATIONS 26

15. Consider the non-linear stochastic differential equation

dXt = rXt (K − Xt ) dt + βXt dBt , X0 = x > 0.

Comparing to the deterministic Bernoulli equation, do a substitution Yt = Xt−1 , then

dYt = −rYt (K − Xt ) dt − βYt dBt + β 2 Yt dt

= (−rK + β 2 )Yt dt − βYt dBt + r dt.

Next do a new change of variables

2 )t
Zt = Yt e(rK−β

and calculate
2
dZt = −βZt dBt + re(rk−β )t dt
 Z t 
−βBt −1 (rk−β 2 )s+βBs
=⇒ Zt = e x +r e ds .
0

Conclude that
2 )t 2
e(rk−β e(rk−β )t+βBt
Xt = = Rt .
Zt x−1 + r 0 e(rk−β 2 )s+βBs ds

16. Consider the non-linear stochastic differential equation

dXt = f (t, Xt ) dt + c(t)Xt dBt , X0 = x.

 R 
t Rt
(a) Let Ft (ω) = exp − 0 c(s) dBs + 21 0 c(s)2 ds . Then calculate

d(Ft Xt ) = Xt dFt + Ft dXt + d[Ft , Xt ]

  
1 2 1 2
= Xt Ft −c(t) dBt − c(t) dt − c(t) dt
2 2
+ [f (t, Xt )Ft dt + c(t)Xt Ft dBt ] − c(t)2 Ft Xt dt
= f (t, Xt )Ft dt.

(b) Defining Yt = Ft Xt , deduce that

dYt
= Ft (ω)f (t, Ft−1 (ω)Yt (ω))
dt
(c) Consider dXt = Xt−1 + αXt dBt , X0 = x > 0. Then
dYt 2
= e−2αBt +α t Yt−1 ,
dt
CHAPTER 5. STOCHASTIC DIFFERENTIAL EQUATIONS 27

which implies
s Z t
Yt = Y02 +2 e−2αBt +α2 s ds
0

and
s
2
Z t
αBt − α2 t
Xt = e x2 +2 e−2αBt +α2 s ds.
0

(d) Consider dXt = Xtγ dt + αXt dBt , X0 = x > 0. Then

dYt α2
= e−(1−γ)Bt +(1−γ) 2 t Ytγ ,
dt
which implies
1
 Z t 2
 1−γ
−(1−γ)Bs +(1−γ) α2 s
Yt = Y01−γ + (1 − γ) e ds
0

and
 Z t 1
 1−γ
2 2
αBt − α2 t 1−γ −(1−γ)Bs +(1−γ) α2 s
Xt = e x + (1 − γ) e ds .
0

Rt Rt
17. Let v ≥ 0 satisfy v(t) ≤ C + A 0
v(s) ds and consider quantity w(t) = 0
v(s) ds. Then
Z t
0
w (t) = v(t) ≤ C + A v(s) ds = C + Aw(t).
0

Then for f (t) = w(t)e−At , calculate

f 0 (t) = e−At (w0 (t) − Aw(t)) ≤ Ce−At

and
Z t
−At C
w(t)e ≤ Ce−As ds = (1 − e−At )
0 A
C At
=⇒ w(t) ≤ (e − 1).
A
Deduce that

v(t) ≤ C + Aw(t) ≤ CeAt .

Chapter 6

The Filtering Problem

28
Chapter 7

Diffusions: Basic Properties

29
Chapter 8

Other Topics in Diffusion Theory

30
Chapter 9

Applications to Boundary Value Problems

31
Chapter 10

Applications to Optimal Stopping

32
Chapter 11

Applications to Stochastic Control

33
Chapter 12

Applications to Mathematical Finance

34