Well-Posedness and Comparison Principle for

Option Pricing with Switching Liquidity

arXiv:1502.07622v1 [q-fin.MF] 26 Feb 2015

T.B. Gyulov and L.G. Valkov

University of Ruse Angel Kanchev, 7017 Ruse, Bulgaria

February 27, 2015

Abstract
We consider an integro-differential equation derived from a system of
coupled parabolic PDE and an ODE which describes an European option
pricing with liquidity shocks. We study the well-posedness and prove
comparison principle for the corresponding initial value problem.

Introduction

This work is devoted to the study of an initial value problem of the following
form



Z
u 1 2 2 2 u
u(S,s)
h(S)
u(S, )
+ ,
=

S
e
ds
+
e
01
10

2
S 2
0

u (S, 0) = h(S).
(1)
Here [0, T ], S (0, +), h(S) is a given function and , 01 , 10 , and
are constants.
The integro-differential equation in (1) is derived from a system of coupled
parabolic PDE and ODE which is suggested by M. Ludkovski and Q. Shen [6] in
European option pricing in a financial market switching between two states -a
liquid state (0) and an illiquid (1) one. We briefly describe their model. First, it
is assumed that the dynamics of the liquidity is represented by a continuous-time
Markov chain (Mt ) with intensity rates of the transitions 0 1 and 1 0 and
determined by the constants 01 and 10 , respectively. During the liquid phase
(Mt = 0) the market dynamics follows the classical Black-Scholes model. More
precisely, the price St of a stock is modelled by geometric Brownian motion
dSt = St dt + St dWt ,
with drift and volatility and a standard one-dimensional Brownian motion
(Wt ) which is independent of the Markov chain (Mt ) (under the real world
probability P). Then the wealth process (Xt ) satisfies
dXt = t Xt dt + t Xt dWt ,
e-mails:

tgulov@uni-ruse.bg (T.Gyulov) and lvalkov@uni-ruse.bg (L.Valkov)

where t denotes the proportion of stock holdings in the total wealth Xt . For
simplicity, it is assumed that the interest rate of the riskless asset is zero.
Respectively, in the illiquid phase (Mt = 1), the market is static and trading
in stock is not permitted, i.e., dSt = dXt = 0.
The presence of liquidity shocks is a source of non-traded risk and makes
the market incomplete. Ludkovski and Shen investigate expected utility maximization with exponential utility function:
u(x) = ex,
where > 0 is the investors risk aversion parameter. The value functions
i (t, X, S), i = 0, 1 for the optimal investment problem are defined as follows:
U
h
i
i (t, X, S) := sup EPt,X,S,i e(XT +h(ST )) , i = 0, 1,
U
t

where EPt,X,S,i is the expectation under the measure P with starting values St =
S, Xt = X and Mt = i. The supremum above is taken over all admissible
trading strategies (t ) and the function h(S) denotes the terminal payoff of a
contingent claim. Standard stochastic control methods and the properties of
the exponential utility function imply that the value functions can be presented
by
i (t, X, S) = eX eRi (t,S) , i = 0, 1,
U
where Ri (t, S) are the unique viscosity solutions of the system ([6])
(
1
0
0
01
01 e(R R ) + d0 +
= 0,
Rt0 + 12 2 S 2 RSS

10 (R0 R1 )
10
1
Rt e
+ = 0,

(2)

with the terminal condition Ri (T, S) = h(S), i = 0, 1. Here d0 := 2 /2 2 .

Let p and q denote the buyers indifference prices corresponding to liquid
0 (t, X
and illiquid initial state respectively. They are defined as follows: U
0
1
1
i
(t, X q, S) = V (t, X) where V , i = 0, 1 are the
p, S) = V (t, X) and U
value functions of the Merton optimal investment problem (i.e. the case when
h(S) 0). It can be shown that p and q satisfy a system of differential equations
which is quite similar to (2) (see (15)). In fact,
p = R0 + 1 ln F0 (t)

and

q = R1 + 1 ln F1 (t)

where
F0 (t) = c1 e1 t + c2 e2 t


1
c1 (d0 + 01 1 ) e1 t + c2 (d0 + 01 2 ) e2 t
F1 (t) =
01
1,2 =
c1 =

d0 + 01 + 10
2 d0 1 T
e
,
2 1

q
2
(d0 + 01 + 10 ) 4d0 10
2

and

c2 =

1 d0 2 T
e
.
1 2

Indifference pricing was first used in the pioneering paper of Hodges and
Neuberger [3]. We refer also to [2] for further applications (see [4] and [8] as
well).
The existence of classical solutions was proved in [6] when the payoff function
h(S) is bounded. This case is restrictive since it does not include such typical
example as the call option h = max {S K, 0} with strike price K. We investigate the solvability of the problem and prove the existence and uniqueness of
a weak solution in suitable Sobolev weighted spaces which allows unbounded
terminal payoff functions.
The integro-differential equation (1) is derived from (2) as follows. Denote
r0 := R0 , r1 = R1 . The system of differential equations for r0 and r1 has the
following from:

r0 1 2 S 2 r0 = e(r1 r0 ) + d +
01
0
01
SS

(3)
2
r1 = e(r0 r1 ) +
10
10

where = T t. The ODE in (3) can be solved explicitly with respect to

r1 . Then we obtain the initial value problem (1) under the substitution u :=
r0 10 and := d0 + 01 10 .
The paper is organized as follows. In Section 2 we prove a comparison
principle (Theorem 2.1) for classical solutions to the problem (1). Then, in
Section 3 we prove a comparison principle (Theorem 3.4) for weak sub/super
solutions. In addition, we study the existence and uniqueness of weak solutions
in a suitable weighted Sobolev space (see Theorem 3.7).

Comparison principle for classical solutions

In this section we consider solutions of (1) satisfying

|u| , |h| A exp ln2 S = AS ln S ,

(4)

for some positive constants A and . Note that conditions (4) include for
example linear growth, polinomial and powers of S with arbitrary exponent.
We prove the following comparison principle:

Theorem 2.1. Let u1 , u0 C((0, +) [0, T )) C 2,1 ((0, +) (0, T )) be two

clasical solutions of (1) corresponding to the initial data h = h1 and h = h0 ,
respectively and such that the conditions (4) hold. Then
inf (h1 h0 ) u1 u0 sup (h1 h0 ) .

(5)

We will only prove the lower bound in (5) since the upper one follows immediately from it. In addition, we can assume that
h := inf (h1 h0 ) > ,
otherwise the left inequality in (5) is trivial. We will use the following auxiliary
lemma

Lemma 2.2. Let u1 and u0 be as in Theorem 2.1 and 1 0 be such that

u1 (S, ) u0 (S, ) h for any [0, 1 ]. Then, there exists a constant > 0
such that u1 (S, ) u0 (S, ) h for any [0, 1 + ). In addition, depends
only on defined in (4) and .
Proof. Let u1 and u0 be two solutions of (1) corresponding to the initial conditions u1 (S, 0) = h1 (S) and u0 (S, 0) = h0 (S). Denote u = u1 u0 ,
= (h1 h0 ), u = u1 + (1 ) u0 , h = h1 + (1 ) h0 , for [0, 1]
h
and define


Z
u(s)
g
u( )
+ .
10
e
ds + e
F [ ; u, g] := 01 e
0

Then
Z 1
d
F [ ; u1 , h1 ] F [ ; u0 , h0 ] =
(F [ ; u , h ]) d
0 d


Z 1
Z
= 01 u

eu ( ) 10
eu (s) ds + eh d
0

+ 01

(7)

u ( )


Z
10

u (s)

u
(s) ds + e

h d

= 01 10

(6)

eu ( )u (s) (
u ( ) u (s)) dsd


Z
01 u ( ) h

(8)

eu ( )h d

and
Z
Z 1
1
(
u ( ) u (s)) ds
eu ( )u (s) d
u
2 S 2 uSS = 01 10
2
0
0

Z 1

( ) h
01 u
eu ( )h d

(9)

Next, define
ln S 12 2 (T1 )
2 2 (T1 )

exp
(S, ) :=
T1

2 !

(10)

where T1 > 0 and (S, ) (0, +) [0, T1 ). Note that LBS =

1 2 2
2 S SS2 =
0 and is increasing with respect to in the interval
T1 4/ , T1 . Choose T1 > 1 in (10) such that the inequality
<

1
,
2 2 (T1 )

holds for all [1 , T1 ) and T1 4/ 2 < 1 . It is enough to define T1 := 1 + ,

where 0 < < min

o

1
+ . Then
2 2
, 4/ 2 . Next, let = u

Z
Z 1
1
( ) 2 S 2 ( )SS = 01 10
(
u ( ) u (s)) ds
eu ( )u (s) d
2
0
0

Z 1

( ) h
01 u
eu ( )h d
(11)
0

01 10 (
u ( ) h)
01 10

ds

(12)

(
u ( ) u
(s)) ds


Z

( ) h
01 u

eu ( )u (s) d
Z

eu ( )u (s) d

eu ( )h d

We will prove that h for any [1 , T1 ). Indeed, assume by contradiction

that inf < h. Note that | =1 > h and there exist S and S such that > h
In fact, + uniformly when either |ln S| +
if either S S or S S.
or T1 . The last observations imply that attains minimum in an interior
(1 , T1 ) and (S , ) < h. Then, ( ) (S , ) = 0,
point (S , ) (S, S)

( )SS (S , ) 0 and
u
u(S , ) h
(S , ) h = (S , ) h (S , ) < 0

(13)

s [1 , ],

(14)

u
(S , ) u
(S , s) = (S , ) (S , s)
((S , ) (S , s)) < 0,

since is increasing in . Thus the right hand side of (12) is positive, a contradiction. Hence = u
+ h for any [1 , T1 ). Let 0. Then
u
= u1 u0 h for any [1 , T1 ).
Proof. (of Theorem 2.1) The comparison principle follows by induction and the
auxiliary Lemma 2.2: we first take 1 = 0 and prove it in the interval [0, 1/2
],
then let 1 = 1/2
and consider the interval [1/2
, ] and etc.
Now, as a corollary we formulate comparison principle for the buyers indifference prices p(S, t), q(S, t) which satisfy the terminal value problem

01 F1 (qp) d0 + 01
1 F0
1

e
+
=0

pt + 2 S 2 pSS

2
F0

F0

10 F0 (pq) 10
1 F1
qt
e
+
=0

F1
1

p(S, T ) = q(S, T ) = h(S).

(15)

By classical solutions of (15) we mean functions such that p C((0, +)

(0, T ])C 2,1 ((0, +)(0, T )), q C((0, +)(0, T ]), qt C((0, +)(0, T )).
Note that
p = 10 (T t) + ln F0 (t) + u(S, T t),
!
Z T t
u(S,s)
h(S)
,
q = 10 (T t) + ln F1 (t) ln 10
e
ds + e
0

(16)
(17)





since p(t) = 1 r0 + ln F0 (t) and q(t) = 1 r1 + ln F1 (t) . Then, a comparison principle in (p, q) solutions will be equivalent to a comparison principle
for the (r0 , r1 ) variables.
We consider growth conditions analogous to (4)


|p| , |h| A exp ln2 S = AS ln S ,
(18)

for some positive constants A and .

Corollary 2.3. Let (p1 , q1 ) and (p0 , q0 ) be two classical solutions of the system
(15) corresponding to terminal data h h1 (S) and h h0 (S), respectively. If
there exist some positive constants A and such that pi (S, t) and hi (S), i = 0, 1
satisfy the conditions (18), then
inf (h1 h0 ) p1 (S, t) p0 (S, t) sup (h1 h0 ) ,
inf (h1 h0 ) q1 (S, t) q0 (S, t) sup (h1 h0 ) .

(19)
(20)

In particular, let h(S) be bounded from below (or from above) by a constant,
i.e. h(S) h (resp. h(S) h ) and p(S, t), q(S, t), be a classical solutions of
the terminal value problem (15) satisfying (18). Then
p(S, t) h and q(S, t) h (respectively p(S, t) h and q(S, t) h ),
for any S (0, +) and any t (0, T ].
Proof. The inequalities (19) follow immediately from Theorem 2.1 and representation (16). In order to prove (20) we will use (17), i.e.
!#
"
Z
T t

qi (, t) = 1 10 (T t) + ln F1 (t) ln 10

eui (,s) ds + ehi ()

for i = 0, 1. Similarly to the proof of Lemma 2.2 we derive

!#
"
Z T t
Z 1
d
d
ln 10
eu (,s) ds + eh ()
q1 (, t) q0 (, t) = 1
0
0 d
R T t u (,s)
Z 1
e
(u1 (, s) u0 (, s)) ds
10 0
1
=
d
R T t u (,s)
0
e
ds + eh ()
10 0
Z 1
eh ()
d
+ (h1 () h0 ())
R T t u (,s)
0 10
e
ds + eh ()
0

Now, (5) implies the estimates (20).

The second part follows immediately due to the fact that p (S, t) h
and q (S, t) h are the solutions of the problem (15) with constant terminal
condition h h . Indeed, if we formally substitute p (S, t) h and q (S, t)
h in (15), then we arrive at the conclusion that it is sufficient to check the
following identities

01 F1
d0 + 01
1 F0
+
= 0,

F0

F0
10
10 F0
1 F1

+
= 0,

F1

F1
6

(21)
(22)

or equivalently
F0 = 01 F1 + (d0 + 01 ) F0 ,

F1

= 10 F0 + 10 F1 ,

(23)
(24)

which follow directly from the definition of F0 and F1 .

Existence of weak solutions

In this section we study the existence and uniqueness of weak solutions in suitable function spaces. First we introduce the weighted L2 space


Z +
2
2
2
Lw := u : kuk0 :=
u (S)w(S)dS < ,
0

given a weight function w > 0. Then we define a weighted Sobolev space as

follows


Hw1 := u : u L2w s.t. Su (S) L2w ,
2

with norm kk1 such that kuk21 = kuk20 + kSu k0 .

Let : [0, +) [0, 1] be increasing, infinitely continuously differentiable
function and such that 0 on [0, 1/2] and 1 on [1, +). We will use to
construct a sequence {u } of compactly supported functions converging in Hw1
to a given element u Hw1 . More precisely, the following auxiliary result holds.
Lemma 3.1. Let (x) := (x/) [1 (x/2)], 0 < < 1 and u := u. Then
u u in Hw1 , as 0.

Proof. Note that (u u ) = (1 ) u u,

S (S) = (S/) (S/) [1 (S/2)] (S/2) (S/2)(S/)
is uniformly bounded with respect to and 1 0 as well as S (S) 0
as 0. Then the Lebesgues dominated convergence theorem implies that
ku u k 0 as 0.
Next, let u(S) be twice continuously differentiable on (0, +) and denote
the operator Lu := 21 2 S 2 u . Then after integration by parts we formally
obtain:
Z
1 2 +
wS 2 u vdS
(Lu, v)L2w =
2
0




Z
1 2 +
w
2

=
wS u v + S
+ 2 wSu v dS,
2
w
0
provided that the integrals above are well-defined, w is continuously differentiable and wS 2 u v as S 0 and S . For example, the above holds
when v is continuously differentiable and with compact support.
Following the above observations we introduce the bilinear form:


 
Z
1 2 +
w

a(u, v) :=
+ 2 v dS.
(25)
wSu Sv + S
2
w
0
7

If the weight function w is twice continuously differentiable, and there exists a

constant C > 0, such that

 

 w (S)   2 w (S) 
S
 , S

(26)
 w(S)   w(S)  C, S (0, +).
then the bilinear form a(u, v) is continuous and semi-coercive on Hw1 , i.e.,
|a (u, v)| c kuk1 kvk1 ,
a (u, u)

2
kuk1

u, v Hw1

2
kuk0

Hw1

(27)
(28)

for some suitable constants c > 0, > 0 and > 0 which are independent of u
and v.
We can choose such weight function that the call option payoff function
h = max {S K, 0} belongs to the space Hw1 , for example, take w := (1 + S) ,
where < 3.
In addition, we assume that
Z +
:=
w(S)dS < +.
(29)
0

This assumption guarantees that any bounded and measurable function belongs
to L2w .
Lemma 3.2. There exists a constant c0 > 0 such that
2

|u(S)| c0 kuk1

1
exp(C |ln S|),
S

u Hw1 ,

(30)

where C satisfies (26).

Proof. Note that there exists a constant c0 such that
2

|u(1)| c0 kuk1 ,

u Hw1 ,

(31)

due to the Sobolev embbeding theorem.

Let S be fixed and denote v() := u(S). We have
2

kvk1 =
=



2
w() 2 S 2 (u (S)) + u2 (S) d



w()
2
w(S) 2 S 2 (u (S)) + u2 (S) d (S)
Sw(S)

1
2
exp(C |ln S|) kuk1 ,
S

since
w()
1
= exp
Sw(S)
S

(32)
(33)
(34)

w ()
d
w ()

Then (30) follows from (31) since v(1) = u(S).

1
exp (C |ln S|) .
S

The space Hw1 is densely and continuously embbeded in L2w . We consider

the Gelfand triples
Hw1 L2w Hw ,
and

L2 (0, T ; Hw1 ) L2 (0, T ; L2w ) L2 (0, T ; Hw ),

where Hw is the dual of Hw1 . Next, we define the set



W (0, T ) := u L2 (0, T ; Hw1 ), u L2 (0, T ; Hw ) ,

(35)

where u is the distributional derivative of u. It is well known (see Lions and

Magenes[5]) that
W (0, T ) C([0, T ], L2w ).

For simplicity we will further write u( ) instead of u(S, ) when this does
not lead to misunderstanding. Recall that


Z
u(s)
h
u( )
+ .
10
e
ds + e
F [ ; u, h] := 01 e
0

Definition 3.3. A function u W (0, T ) is called weak supersolution (subsolution) of the initial value problem (1) if u(0) h (resp. u(0) h) and for
a.a. (0, T ) the inequality
Z +
hu,
vi + a(u, v) ()
wF [ ; u, h] vdS,
(36)
0

Hw1 .

holds for any nonegative v

Respectively, the function u W (0, T ) is
called weak solution of the initial value problem (1) if u(0) = h and for a.a.
(0, T ) the equality
Z +
hu,
vi + a(u, v) =
wF [ ; u, h] vdS,
v Hw1 ,
(37)
0

holds.

Next, we prove the following comparison principle for weak super/subsolutions

satisfying growth conditions of type (4).
Theorem 3.4. Let u be a weak supersolution of the initial value problem (1)
with initial data h(S) h and u be a weak subsolution corresponding to the
initial data h(S) h where h and h are given and h h. Assume in addition,
that there exist positive constants A and such that
 


|h| , h , |u| , |u| A exp ln2 S = AS ln S ,
(38)
for a.a. (S, t) (0, +) [0, T ].
Then u u for a.a. (S, t) (0, +) [0, T ].

Denote u := u u. We will prove that u := max {u, 0} = 0 almost

everywhere. Similarly to (9), we obtain that the following inequality holds for
a.a. (0, T ) and for any nonegative v Hw1 with compact support in (0, +):

Z Z
(, s) (u (S, ) u (S, s)) ds v(S)wdS
hu,
vi + a(u, v) 01 10
0

(39)

01

u (S, ) h(S)
v(S)( )wdS,
9

where
(, s) :=

u ( )u (s)

d,

( ) :=

eu ( )h d,

:= h h 0 and h := h + (1 ) h.
u := u + (1 ) u, u(, 0) h
It is sufficient to prove the following auxiliary result:
Lemma 3.5. Assume that 1 0 is such that for any t [0, 1 ] the inequality
u(t) u(t) 0 holds a.e. on (0, +). Then the same inequality holds for any
t [0, 1 + ], where > 0 is a constant which depends only on and .
Proof. Let be defined by (10) and u := u + where u = u u. Then,
assume that is chosen as in the proof of Lemma 2.2. We will prove that
u := max {u , 0} 0 for a.a. (S, t) (0, +) [1 , 1 + ]. Note that there
exist a closed interval I (0, +) such that u = 0 on the set ((0, +) \ I )
[1 , 1 + ] due to the conditions (38). Now, let (S) be a smooth function with
compact support in (0, +) such that (S) = 1 on the interval I . Then
u L2 (1 , 1 + ; Hw1 ) and (u ) = u . Next, for any nonnegative v Hw1
with compact support supp v I we have v = v, a(u, v) = a(u, v) and then


d
(u ) , v + a (u , v) = hu,
vi + h,
vi + a(u, v) + a(, v) (40)
d
1
= hu,
vi + a(u, v) 2 (2 + , v)L2w
2
{z
}
|
=0

Z Z
01 10
(, s) (u (S, ) u (S, s)) ds v(S)wdS
(41)
0
0
Z 


01
u (S, ) h(S)
v(S)( )wdS
0
Z

Z
1
(, s)ds u (S, ) v(S)wdS
(42)
01 10
0
0
Z

Z

01 10
(, s) (u (S, ) u (S, s)) ds v(S)wdS
1
0
Z
01
u (S, ) v(S)( )wdS,
0

i.e.,




Z Z 1
d
(, s)ds u (S, ) v(S)wdS
(u ) , v + a (u , v) 01 10
d
0
0
(43)

Z Z
01 10
(, s) (u (S, ) u (S, s)) ds v(S)wdS
0
1
Z
01
u (S, ) v(S)( )wdS,
0

where we have used the fact that u > u and u (S, ) u (S, s) > u (S, )
u (S, s) for any s [1 , ] since (S, ) is increasing on that interval. Now, take
v = u and note that u = u+ u , a (u , u ) = a (u , u ) and
u (S, s) u (S, ) u (S, s) u (S, )
10

1 2
u (S, s) + u2 (S, ) .
2

After integration with respect to form 1 to t [1 , 1 + ] the inequality (43)

implies

Z t Z
1
2
2
ku (t)k0 + a (u , u )
(S, )u (S, ) wdS d,
(44)
2
1
0
where
(S, ) := 01 10

Z

(, s)ds +

1
2

(, s)ds

1
2


(s, )ds + 01 ( ).

|(S, )| is bounded from above by a constant, say C > 0, when S I and due
to the semi-coercivity of the bilinear form a(, ) (see (28)) we obtain:
1
2
ku (t)k0 (C + )
2

ku ( )k0 d.

(45)

Hence the Gronwall inequality implies ku (t)k0 = 0 for any t [1 , 1 + ]

since ku (1 )k0 = 0. Then u + 0 a.e. Thus u 0 a.e. since > 0 is
arbitrary.
We further prove another useful estimate.
Lemma 3.6. There exists a constant C > 0 such that


max ku(t)k0 + kukL2 (0,T,H 1 ) C ku(0)k0 + k
ukW (0,T ) + khk0 + 1 (46)
w

t[0,T ]

for any weak subsolution u and any function u W (0, T ) satisfying u u.

Proof. Let v Hw1 be some nonnegative function. We have
hu,
vi + a(u, v)

wF [ ; u, h] vdS,
Z

01 10
[u( ) u(s)] ds , v
0

(47)
L2w

01 (u( ) h , v)L2

+ ( 01 10 01 ) (1 , v)L2w .
Take v = u u
and integrate (47) with respect to from 0 to t.
Z tD
t
E
1
1

2
2
ku(t)k0 + a(u, u) ku(0)k0 + (u, u
)L2w  + a(u, u
u , u d
(48)
)
2
2
0
0
Z
2
Z t

1 t
2

01
u(
)d
(10 + 1) ku( )k0 d + 01 10


2 0
0
0


ukL2 (0,t,L2w ) + khk0 + 1 kukL2 (0,t,L2w )
+ C1 k
+ C2 ( khk0 + 1) k
ukL2 (0,t,L2w ) .

Then a standard argument implies the estimate (46).

11

Now, we prove the existence of weak solutions, provided that h Hw1 . The
proof is based on the lower and upper solution method (cf. [7]). However, the
exponential nonlinearity in (1) causes some very technical difficulties which have
to be overcome.
Theorem 3.7. Assume that h Hw1 . Then there exist a weak solution u to the
initial value problem (1). Moreover, there exists a constant C > 0 independent
of u such that
kuk
L2 (0,T,L2 ) + kukL (0,T,H 1 ) C (ku(0)k1 + 1)
w

(49)

Proof. We will present the proof in several steps.

Step 1. Let h L2w be bounded. Then there exists a weak solution u to the
initial value problem (1). In addition, if u(0) = h Hw1 , then the inequality
(49) holds with a constant C independent of u(0).
Note that we can conctruct appropriate couple of a supersolution u and a
subsolution u. Indeed, let the constant c0 be such that |h| c0 and take
u := c0 M t for some positive constant M . If M is great enough then u is a
subsolution. Analogously, u := c0 + M t is a supersolution provided that M .
Next, according to (8) we can choose a constant N > 0 such that


Z
N u( ) + F [ ; u, h] = N u( ) 01 eu( ) 10
eu(s) ds + eh +
0

is increasing in u, i.e.
N u1 ( ) + F [ ; u1 , h] N u0 ( ) + F [ ; u0 , h] ,
for all u0 and u1 such that u u0 u1 u. Now, we can construct a decreasing
sequence of supersolutions u0 := u, u1 , u2 , ... such that un+1 is the solution of
the initial value problem

u n+1 21 2 S 2 un+1,SS + N un+1 = N un + F [ ; un , h] ,
un+1 (S, 0) = h(S)
and u un u. A standard argument implies that un converges to a weak
solution of the problem (1). We omit the details.
Next, assume in addition that h Hw1 . Then u L2 (0, T ; L2w ) and u

L (0, T ; Hw1 ) (see, e.g., Bonnans [1]) and the following parabolic estimate holds:


kuk
L2 (0,T,L2 ) + kukL (0,T,H 1 ) c0 ku(0)k1 + kF [; u, h]kL2 (0,T,L2 )
w

We will prove the stronger estimate (49). First, we have

1
2
2

1
(50)
2 S 2 uSS = F [ ; u, h] u L2 (0, T, L2w ),
2




1
1
1
2
2
S 2 uSS , u L2 d = 2
ku(t)k1 ku(0)k1
(51)
w
2
2
2


Z  
w
1 2 t

S S
+
+ 2 uS u, u
d
2
w
0
L2
w

12


d
(F [ ; u, h]) d wdS
(F [ ; u, h] , u)
L2w d =
0
0 d

Z + Z t
+
(u + 01 10 ) d wdS
Z

Z

(52)

|| 1/2

ku ( )k0 d + 01 (1 + 10 t)

(53)

since




Z
d
d
eu(s) ds + eh +
01 eu( ) 10
(F [ ; u, h]) =
d
d
0


Z
u( )
u(s)
10
= 01 e
e
ds + eh u 01 10

(54)

= F [ ; u, h] u u 01 10 .

and

t
0

d
(F [ ; u, h]) d = F [t; u, h] F [0; u, h] 01
d

(55)

(56)

We multiply both sides of the equation u 1/2 2 S 2 uSS = F [ ; u, h] with u in

L2w and integrate from 0 to T . Then (51) and (53) imply


Z t
Z  
1 2
w
1 2 t
2
2

kuk
0 d + ku(t)k1
S S
+ 2 uS u, u
d (57)
4
2
w
0
0
L2w
Z t
1
2
+ || 1/2
ku ( )k0 d + 2 ku(0)k1
4
0
+ 01 (1 + 10 t)
Z t

C
(ku( )k1 + 1) ku(
)k0 d + ku(0)k21 + 1
0

for some constant C > 0. Now, a techical, but standard argument implies that
(49) holds.
Step 2. Let h Hw1 be bounded from below, i.e., u(0) = h c. Then
there exists a weak solution u to the initial value problem (1). In addition, the
inequality (49) holds.
Let (x) be defined as in Lemma 3.1, i.e., (x) := (x/) [1 (x/2)]. Step
1 implies that there exists a solution u corresponding to the initial condition
u (0) = (hc)+c = h+(1)c which is bounded. Moreover, h+(1
)c h increases as 0 and converges in Hw1 to h. Then the comparison
principle from Theorem 3.4 implies that the sequence u is increasing as
0. Next, the estimate (49) and Lemma 3.2 imply that u (S, ) converges to
a finite limit u(S, ) for any (S, ) (0, +) [0, T ]. What is more, u is
weakly convergent to u(S,

) in L2 (0, T ; L2w ), u is weakly- convergent to u in

1
L (0, T, Hw ) and u satisfies the estimate (49). Then it is sufficient to prove that
F [ ; u , h + (1 )c] is weakly convergent to F [ ; u, h] in L2 (0, T ; Hw ).
First, note that
1
F [ ; u , h + (1 )c] = u 2 S 2 u,SS
2
13

is bounded in L2 (0, T ; Hw ) and then there exists an element F L2 (0, T ; Hw )

such that
F [ ; u , h + (1 )c] 2 F .
L (0,T ;Hw )

On the other hand, F [ ; u , h + (1 )c] is bounded from above by the constant function . Let v L2 (0, T ; Hw1 ) be some arbitrary nonnegative function.
Then Fatous lemma implies
D
E
v = lim ( F [; u , h + (1 )c] , v) 2
F,
L (0,T ;L2w )
0

h F [; u, h] , vi 0,

i.e.

(58)

F [; u, h] L2 (0, T ; Hw ) and F [; u, h] F .
Finally, we prove that in fact
D
E
F [; u, h] F , i.e., hF [; u, h] , vi = F , v

v L2 (0, T ; Hw1 ).

(59)

First, observe that, v := v v as 0 in L2 (0, T ; Hw1 ). Hence, it is sufficient

to prove (59) for functions v vanishing outside a set of the form I [0, T ] where
I (0, +) is some closed interval. According to estimate (49) and Lemma
3.2 (applied to the interval I) the functions u and u are uniformly bounded on
I [0, T ]. Then
E
D
hF [; u, h] , vi = lim (F [; u , h + (1 )c] , v)L2 (0,T ;L2w ) = F , v .
0

Step 3. Let h Hw1 . Then there exists a weak solution u to the initial value
problem (1). In addition, the inequality (49) holds.
Consider a sequence of problems with initial condition
uN (S, 0) = max {h(S), N } ,

N = 1, 2, . . . .

Then the corresponding solutions uN form a decreasing sequence due to the comparison principle and Lemma 3.2. Moreover, the pointwise limit limN uN (S, )
is finite for any (S, ) since the inequality (49) holds for each function uN . Then
the proof follows similar arguments as in Step 2.
Finally, note that the uniqueness of the weak solution is a consequence of
the comparison principle. More precisely, we have the following corollary.
Corollary 3.8. Assume that h Hw1 . Then there exists a unique weak solution
u W (0, T ) L (0, T, Hw1 ) to the initial value problem (1). Moreover, the
estimate (49) holds with a constant C > 0 independent of u.
Acknowledgement
The research is supported by the European Union under Grant Agreement
number 304617 (FP7 Marie Curie Action Project Multi-INT STRIKE - Novel
Methods in Computational Finance). The second author is also supported by
Bulgarian National Fund of Science under Project I02/20-2014.

14

References
[1] J.F. Bonnans, Numerical analysis of partial differential equations
arising in finance, Master 2 Probabilites et Finance, Paris VI
and Ecole Polytechnique: revision of Aug. 29, 2013, available at
http://www.cmap.polytechnique.fr/bonnans/notes/edpfin/edpfin.html
[2] R. Carmona (Ed.), Indifference Pricing: Theory and Applications, Princeton University Press, Princeton NJ, 2008.
[3] S.D. Hodges, A. Neuberger, Optimal Replication of Contingent Claims under Transaction Costs, Rev. Futures Markets 8, (1989), 222239.
[4] T.S.T. Leung, A Markov-modulated stochastic control problem with optimal multiple stopping with application to finance, Decision and Control
(CDC), 49th IEEE Conference, IEEE, (2010), 559-566.
[5] J. L. Lions, E. Magenes, Probl`emes aux Limites non Homog`enes et Applications, Vols. I and II., Dunod, Paris, 1968.
[6] M. Ludkovski, Q. Shen, European option pricing with liquidity shocks, Int.
J. Theor. Appl. Finan., 16, No. 7, Article ID 1350043, 30 p. (2013). ISSN
0219-0249
[7] C. V. Pao, Nonlinear parabolic and elliptic equations. Plenum Press, New
York, 1992.
[8] T. Zhou, Indifference valuation of mortgage-backed securities in the presence of prepayment risk, Mathematical Finance 20 (2010), no. 3, 479-507.

15