International Journal of Computational Intelligence Systems, Vol.2, No.

1 (March, 2009), 83-90

Aumann Type Set-valued Lebesgue Integral

and Representation Theorem

Jungang Li, Shoumei Li

Department of Applied Mathematics, Beijing University of Technology,
100 Pingleyuan, Chaoyang District,
Beijing, 100124, P.R.China
E-mail: [email protected], [email protected]

Received: 31-05-2008
Revised: 29-01-2009

Abstract
In this paper, we shall firstly illustrate why we should discuss the Aumann type set-valued Lebesgue
integral of a set-valued stochastic process with respect to time t under the condition that the set-valued
stochastic process takes nonempty compact subset of d-dimensional Euclidean space. After recalling
some basic results about set-valued stochastic processes, we shall secondly prove that the Aumann type
set-valued Lebesgue integral of a set-valued stochastic process above is a set-valued stochastic process.
Finally we shall give the representation theorem, and prove an important inequality of the Aumann type
set-valued Lebesgue integrals of set-valued stochastic processes with respect to t, which are useful to
study set-valued stochastic differential inclusions with applications in finance.

Keywords: set-valued stochastic process, set-valued Lebesgue integral, Aumann type integral, represen-
tation theorem.

1. Introduction the Aumann type Lebesgue integral of the set-

valued
R stochastic process F with respect to time
In studying the evolution of macro-systems in eco- τ , st Gτ (xτ )dBτ is the Aumann type Ito integral of
nomic, social or biological sciences, the dynam- the set-valued stochastic process G with respect to
ical systems having velocities are not determined the Brownian motion B. It appears in many prob-
uniquely by the state of systems. Thus, we study lems, for instance, it can be considered in a natural
the differential inclusion instead of differential equa- way as a theoretical description of stochastic control
tion. A stochastic differential inclusion is defined as problems1 .
dxt ∈ Ft (xt )dt + Gt (xt )dBt , x0 = ξ , There are many related former works about
set-valued Lebesgue integral. Based on the work
which can be written in stochastic integral form as of Richter2 and Kudo3 , Aumann4 introduced the
³Z t ´ Lebesgue integral of a set-valued function and dis-
xt −xs ∈ clL2 Fτ (xτ )d τ +Gτ (xτ )dBτ , s,t ∈ I, cussed its properties. Kisielewicz5 introduced the
s
Aumann type Lebesgue integral of a set-valued
where F, G are set-valued stochastic
R processes, stochastic process. Kisielewicz with his colleagues1
B = (Bt )t∈I is a Brownian motion, st Fτ (xτ )d τ is 5 -9 did a lot of nice works about stochastic differ-

Published by Atlantis Press

Copyright: the authors
83
J. Li, S. Li

ential inclusions, especially they discussed solution and prove that the Aumann type Lebesgue integral is
problems. We naturally expect that the Aumann type a set-valued stochastic process and other properties
Lebesgue integral of a set-valued stochastic process of the integral, especially representation theorem of
is a set-valued stochastic process, which is useful for this type integral and an important inequality, which
applications. If F takes nonempty closed set-values, are useful in the study of set-valued stochastic differ-
we can not prove it directly, but by taking decompos- ential inclusions. Finally we shall give an example
able closure. Li and Li10 gave the definition of the for its application in Finance and show conclusions
Lebesgue integral of a set-valued stochastic process and acknowledgement.
by decomposable closure and discussed more prop-
erties of the integral. We also would like to refer to
2. Set-valued Stochastic Processes
related works such as Ref.11-14 and so on.
However, a set-valued stochastic process usually Throughout this paper, assume that (Ω, A , µ ) is a
takes compact subset of Rd space (d-dimensional complete atomless probability space, the σ -field fil-
Euclidean space) in the real world. For example, in tration {A t : t ∈ I} satisfies the usual conditions (i.e.
the famous Black-Scholes formula for the price of a containing all null sets, non-decreasing and right
European call option, the stock price st at the time t continuous). R is the set of all real numbers, N is the
is assumed to satisfy set of all natural numbers, Rd is the d-dimensional
dst = st (udt + vdBt ) Euclidean space with usual norm k · k, B(E) is the
Borel field of the metric space E. Let f = { f (t), A t :
where s0 > 0, u, v are constants, u is the drift of
t ∈ I} be a Rd -valued adapted stochastic process. It
stock, v is the volatility of stock and Bt is a Brown-
is said that f is progressively measurable if for any
ian motion. However, being incomplete or vague of
t ∈ I, the mapping (s, ω ) 7→ f (s, ω ) from [0,t] × Ω
information, one usually predicts the drift of stock
to Rd is B([0,t]) × At -measurable.
within some bounded interval, for example, [u1 , u2 ],
Each right continuous (left continuous) adapted
u1 < u2 , rather than an exact number or an un-
process is progressively measurable.
bounded interval, for the unbounded interval usually
Assume that L p (Rd ) denotes the set of Rd -
has no actual sense. Similarly for the volatility of
valued stochastic processes f = { f (t), A t : t ∈ I}
stock. This becomes a set-valued stochastic differ-
such that f satisfying (i) f is progressively measur-
ential inclusion as follows:
able; and (ii)
dst ∈ st (Ut dt +Vt dBt ),
h ³Z T ´i1/p
where Ut ,Vt are set-valued stochastic processes tak- ||| f ||| p = E k f (t, ω )k p ds < ∞.
ing compact subsets of R as their values. Under the 0

condition that a set-valued stochastic process takes

Let f , f 0 ∈ L p (Rd ), f = f 0 if and only if ||| f −
nonempty compact subset of Rd , can we prove that p d
f 0 |||
p = 0. Then (L (R ), ||| · ||| p ) is complete.
the Aumann type set-valued Lebesgue integral is a
Now we review notation and concepts of set-
set-valued stochastic process? What properties does
valued stochastic processes.
the integral have? These are the problems we shall
solve in this paper. Fortunately, we also find an al- Assume that K(Rd ) is the family of all nonempty,
most everywhere problem in the former definition closed subsets of Rd , and Kc (Rd ) (resp. Kk (Rd ),
of set-valued Aumann type Lebesgue integral, and Kkc (Rd )) is the family of all nonempty closed con-
shall solve it. vex (resp. compact, compact convex) subsets of Rd .
We organize our paper as following: in Section For any x ∈ Rd , A is a nonempty subset of Rd , define
2, we shall introduce some necessary notations, def- the distance of x and A as d(x, A) = infy∈A kx − yk.
initions and results about set-valued stochastic pro- The Hausdorff metric on K(Rd ) is defined as
cesses. In Section 3, we shall discuss the former def-
dH (A, B) = max{sup d(a, B), sup d(b, A)}
inition of Aumann type set-valued Lebesgue integral a∈A b∈B

Published by Atlantis Press

Copyright: the authors
84
 Aumann Type Lebesgue Integral

for A, B ∈ K(Rd ). For B ∈ K(Rd ), define kBkK = stochastic processes. Similarly, we have notations
dH ({0}, B) = supa∈B kak. L p (Kc (Rd )), L p (Kk (Rd )) and L p (Kkc (Rd )).
Take Fi = {Fi (t) : t ∈ I} ∈ L p (K(Rd )), i = 1, 2, de-
For a set-valued random variable F(Ref. 15, 16), fine
define the set h ³Z T ´i1/p
N p (F1 , F2 ) = E dHp (F1 (s, ω ), F2 (s, ω ))ds .
SFp = { f ∈ L p [Ω; Rd ] : f (ω ) ∈ F(ω ) for a.e. ω ∈ Ω}, 0
F1 and F2 are said to be equivalent, if
where L p [Ω; Rd ] is the set of all Rd -valued random
N p (F1 , F2 ) = 0, denoted by F1 = F2 . We have
variables f such that k f k p = [E(k f k p )]1/p < ∞, and
that (L p (K(Rd )), N p ) is complete, L p (Kc (Rd )),
constant p > 1. The expectation of F is defined as
L p (Kk (Rd )) and L p (Kkc (Rd )) are closed sub-
E[F] = {E[ f ] : f ∈ SF1 }. It is called Aumann in-
sets of (L p (K(Rd )), N p ). Denote |||F||| p =
tegral introduced by Aumann4 in 1965 . A set- h ³R ´i1/p
p
valued random variable F : Ω → K(Rd ) is called in- E 0T kF(s)kK ds .
tegrable if SRF1 is non-empty. F is called integrable
bounded if Ω kF(ω )kK d µ < ∞. Let L p [Ω; K(Rd )] Theorem 1. Let F ∈ L p (K(Rd )) with p > 1, then
(resp. L p [Ω; Kc (Rd )], L p [Ω; Kkc (Rd )]) denote the S1 (F) = S p (F).
family of K(Rd )-valued (resp. Kc (Rd ), Kkc (Rd )- Proof. S p (F) ⊆ S1 (F) is obvious. Now we
valued) L p -bounded random variables F such that prove the converse. For any f ∈ S1 (F), we have
kF(·)kK ∈ L p [Ω; R]. For any two set-valued random || f (s, ω )|| 6k F(s, ω ) kK since f (s, ω ) ∈ F(s, ω ) for
variables F1 , F2 ∈ L p [Ω; K(Rd )], define a.e. (t, ω ) ∈ I × Ω. Note that F ∈ L p (K(Rd )), so
³Z ´1/p that we have f ∈ L p (Rd ), which implies S1 (F) ⊆
∆ p (F1 , F2 ) = dHp (F1 (ω ), F2 (ω ))d µ , S p (F).
Ω

then (L p [Ω; K(Rd )], ∆ p ) is a complete space. Con- 3. Aumann Type Set-valued Lebesgue Integral
cerning more definitions and more results of set- and its Properties
valued random variables, readers could refer to the
excellent paper 15 or the book 16 . Now we give the definition of Aumann type
Definition 1. A set-valued stochastic process F = Lebesgue integral of a set-valued stochastic process
{F(t) : t ∈ I} is called progressively measurable, with respect to time t.
if for any A ∈ B(Rd ) and any t ∈ I, {(s, ω ) ∈ Definition 3. Let a set-valued stochastic process
[0,t] × Ω : F(s, ω ) ∩ A 6= 0}
/ ∈ B([0,t]) × A t . F F = {F(t) : t ∈ I} ∈ L p (K(Rd )) (1 6 p < +∞). For
is called L p -bounded, if the real stochastic process any t ∈ I, ω ∈ Ω, define
{kF(t)kK , A t : t ∈ I} ∈ L p (R). Z t ½Z t ¾
p
Definition 2. A Rd -valued stochastic process (A) F(s, ω )ds := f (s, ω )ds : f ∈ S (F) ,
0 0
{ f (t), A t : t ∈ I} ∈ L p (Rd ) is called an L p - Rt
selection of F = {F(t), A t : t ∈ I} if f (t, ω ) ∈ where
R 0 f (s, ω )ds is the Lebesgue integral.
F(t, ω ) for a.e. (t, ω ) ∈ I × Ω. (A) 0t F(s, ω )ds is called the Aumann type
Let S p ({F(·)}) or S p (F) denote the family of all Lebesgue integral of the set-valued stochastic pro-
p
L -selections of F = {F(t), A t : t ∈ I}, i.e. cess F with respect to time t introduced in Ref.5.
n For any 0 6 u < t < T ,
S p (F) = { f (t) : t ∈ I} ∈ L p (Rd ) : f (t, ω ) ∈ Z t Z t
o (A) F(s, ω )ds := (A) I[u,t] (s)F(s, ω )ds.
F(t, ω ), for a.e. (t, ω ) ∈ I × Ω . u 0

Let L p (K(Rd )) denote the set of all L p - Remark 1. (1) In the definition 3, the set of se-
bounded progressively measurable K(Rd )-valued lections is S p (F). As a matter of fact, if we only

Published by Atlantis Press

Copyright: the authors
85
J. Li, S. Li

consider the Lebesgue integral, we can use S1 (F). Proof. When p = 1, we have
But we often consider the sum of the Lebesgue inte- Z t
gral of a set-valued stochastic process with respect to Lt (F)(ω ) = (A) F(s, ω )ds
time t and the Ito integral of a set-valued stochastic ½Z t 0 ¾
process with respect to the Brownian motion, where 1
= f (s, ω )ds : f ∈ S (F) .
we have to use S2 (F). Thus we here use S p (F) for 0
R
more general case. From Theorem II.3.20 in Ref.6, (A) 0t F(s, ω )ds
(2) There is a delicate problem in the defini- takes nonempty compact and convex subsets of Rd
tion above, i.e. is the Aumann type Lebesgue in- as its values and we have
tegral of a stochastic process well-defined for every Z t Z t
ω ∈ Ω? As matter of fact, take an f ∈ S p (F). Then, (A) F(s, ω )ds = (A) coF(s, ω )ds.
0 0
for any fixed t ∈ I, by Fubini Theorem, the map-
ping f (·, ω ) : [0,t] → Rd is B([0, t])-measurable for Since F : I × Ω → Kk (Rd ) is progressively measur-
all ω ∈ Ω, but ONLY for a.e. ω ∈ Ω (NOT every able, by Remark II.3.5 in Ref.6, I × Ω 3 (t, ω ) →
ω ∈ Ω!), σ (x, F(t, ω )) ∈ R is measurable for every x ∈ Rd ,
Z t where σ (x, A) = sup{hx, yi : y ∈ A} for A ⊂ Rd . By
It ( f )(ω ) = f (s, ω )ds < ∞. virtue of Theorem II.3.21 in Ref.6, we have
0 Z t ³ ´ ³ Z t ´
σ x, F(s, ω ) ds = σ x, (A) F(s, ω )ds
Now the problem appears: It ( f ) is defined a.e. ω ∈ 0 0
Ω for each f and the set {It ( f ) : f ∈ S p (F)} is usu-
for Revery x ∈ Rd , ω ∈ Ω. So Lt (F)(ω ) =
ally uncountable. We should notice that the union of
(A) 0t F(s, ω )ds is measurable by Theorem II.3.8
the exceptional sets may not be of measure zero. If
in Ref.6
R or Proposition I.2.5 in Ref.18. Thus
not, It (F) is not well-defined even for a.e. ω ∈ Ω!
(A) 0t F(s, ω )ds is measurable when p = 1. Since
Under what kind of condition, is the above It (F)
F ∈ L p (Kk (Rd )), we have SR1 (F) = S p (F) by The-
well-defined for a.e. ω ∈ Ω?
orem 1. So Lt (F)(ω ) = (A) 0t F(s, ω )ds is measur-
To solve the problem, we assume that A is µ -
able with respect to ω ∈ Ω for any p > 1.
separable in this paper. In this case, we have that
S p (F) is separable (Ref.17). Thus the Aumann Remark 2. (1) The Aumann type set-valued
type Lebesgue integral It (F) is well-defined for a.e. Lebesgue integral defined in Theorem 2 is a
ω ∈ Ω. Without loss of generalization, we assume set-valued stochastic process denoted by L(F) =
that for every ω ∈ Ω, definition 3 and the following {Lt (F) : t ∈ I}. Please notice that we proved it under
hold. the condition of the set-valued stochastic process F
Now we prove that the Aumann type set-valued taking nonempty COMPACT set values.
Lebesgue integral is a stochastic process. (2) We are interested in the set of all selections
Theorem 2. Assume that a set-valued stochastic of the integral stochastic process L(F).R For any fixed
process F ∈ L p (Kk (Rd )). Then the set-valued map- t ∈ I, by Fubini Theorem, It ( f )(ω ) =: 0t f (s, ω )ds is
ping Lt (F) : Ω → Kkc (Rd ) defined by an A t -measurable function with respect R to ω for any
Z t given f ∈ S p (F). Thus It ( f )(·) =: 0t f (s, ·)ds is a se-
Lt (F)(ω ) = (A) F(s, ω )ds lection of Lt (F). By the classical Jensen inequality,
0 we have It ( f ) ∈ L p [Ω, A t , µ ; Rd ]. Thus, {It ( f ) : f ∈
S p (F)} is a non-empty subset of L p [Ω, A t , µ ; Rd ].
is measurable, i.e. Lt (F) is a set-valued random
As a matter of fact, we have the following Theorem.
variable, and
Z t Theorem 3. Assume that a set-valued stochastic
Lt (F)(ω ) = (A) coF(s, ω )ds. process F ∈ L p (Kk (Rd )) and continue to use above
0 notations. Then we have that {It ( f ) : f ∈ S p (F)} is
closed in L p [Ω, A t , µ ; Rd ].

Published by Atlantis Press

Copyright: the authors
86
 Aumann Type Lebesgue Integral

R
Proof. Take a sequence {{ fn (t) : t ∈ R I} : n ∈ N} ⊂ For any t ∈ I, SL1t (F) (A t ) = { 0t f (s)ds : f ∈
S p (F) such that {φn (t) : n ∈ N} =: { 0t fn (s)ds : n ∈ S p (F)}. We only need to prove
N} ⊂ {It ( f ) : f ∈ S p (F)} is a Cauchy sequence in nZ t o nZ t o
L p [Ω, A t , µ ; Rd ]. Since L p [Ω, A t , µ ; Rd ] is com- f (s)ds : f ∈ S p (F) ⊂ cl f n (s)ds : n ∈ N ,
plete, there exists φ (t) ∈ L p [Ω, A t , µ ; Rd ] such that 0 0

where the closure is taken in L1 , since the opposite

E[||φn (t) − φ (t)|| p ]
h° Z t inclusion is obvious.
°pi
=E ° fn (s)ds − φ (t)° Let g ∈ S p (F), then there exists a subse-
0 quence { f ni : i > 1} of { f n : n ∈ N} such that
→ 0, (n → ∞). R 1
(E 0T ||g(t,Rω ) − f ni (t, ωR )|| p dt) p → 0(i → ∞). We
know that 0t g(s)ds ∈ { 0t f (s)ds : f ∈ S p (F)}, and
Thus, there exists a subsequence { fnk : k ∈ N} of
Z t Z t
{ fn : n ∈ N} such that when k → ∞, we have
E|| g(s, ω )ds − f ni (s, ω )ds||
Z t 0 0
Z t
fnk (s, ω )ds → φ (t, ω ), a.e. ω ∈ Ω.
0 6E ||g(s, ω ) − f ni (s, ω )||ds
0
Z T
By Theorem 2, Lt (F)(ω ) is a compact subset of
R 6E ||g(s, ω ) − f ni (s, ω )||ds
Rd . This, with the fact It ( fnk )(ω ) = 0t fnk (s, ω )ds ∈ 0
Lt (F)(ω ), implies → 0 (i → 0),
R R
φ (t, ω ) ∈ Lt (F)(ω ), a.e. ω ∈ Ω. which means 0t g(s)ds ∈ cl{ 0t f n (s)ds : n ∈ N}.
Therefore,
Therefore, {It ( f ) : f ∈ S p (F)} is a closed subset of nZ t o nZ t o
L p [Ω, A t , µ ; Rd ]. f (s)ds : f ∈ S p (F) ⊂ cl f n (s)ds : n ∈ N .
0 0
On the other hand, from Theorem 2, we know
that {It ( f ) : f ∈ S p (F)} is decomposable (Ref.15). The proof is completed.
Therefore, SL1t (F) (A t ) = {It ( f ) : f ∈ S p (F)} from Theorem 5. (Representation Theorem) For any
Theorem 3. set-valued stochastic process F ∈ L p (Kk (Rd )),
The following Lemma and Theorem are about there exists a sequence { f i = { f i (t, ω ) : t ∈ I, ω ∈
representation theorem of the Aumann type set- Ω} : i ∈ N} ⊂ S p (F) such that for a.e. (t, ω ) ∈ I ×Ω,
valued Lebesgue integral.
F(t, ω ) = cl{ f i (t, ω ) : i ∈ N},
Lemma 4. If the set-valued stochastic process F ∈
L p (Kk (Rd )), then there exists a sequence { f n : n ∈ and
N} ⊂ S p (F) such that for any t ∈ I, nZ t o
Lt (F)(ω ) = cl f i (s, ω )ds : i ∈ N .
nZ t 0
SL1t (F) (A t ) = cl f n (s)ds : n ∈ N},
0
Proof. For F ∈ L p (Kk (Rd )), by Lemma 4, there
where the closure is taken in L1 . exists a sequence {hn = {hn (t) : t ∈ I} : n ∈ N} ⊂
Proof. Since we assume that A is µ -separable, S p (F) such that for any t ∈ I,
we have that L p (Rd ) is separable (Ref.17). Thus nZ t o
S p (F) is also separable since it is a closed subset SL1t (F) (A t ) = cl hn (s)ds : n ∈ N ,
0
of L p (Rd ) (Theorem 2.4 in Ref.10). That is, there
exists a sequence { f n : n ∈ N} ⊂ S p (F) such that where the closure is taken in L1 . By Theo-
S p (F) = cl{ f n : n ∈ N}, where the closure is taken remR 1.3.1 in Ref.16, for any ω ∈ Ω, Lt (F)(ω ) =
in L p (Rd ). cl{ 0t hn (s, ω )ds : n ∈ N}. Due to Theorem 2.5 in

Published by Atlantis Press

Copyright: the authors
87
J. Li, S. Li

Ref.10, there exits a sequence {φ j = {φ j (t) : t ∈ I} : Proof. For each f ∈ S p (F) and for any ω ∈ Ω, we
j ∈ N} ⊂ S p (F) such that for any (t, ω ) ∈ I × Ω, have

F(t, ω ) = cl{φ j (t, ω ) : j ∈ N}.

inf kIt ( f )(ω ) − yk2
y∈Lt (G)(ω )
Take the element one by one from two sequences °Z t Z t °2
{hn : n ∈ N}, {φ j : j ∈ N}, and get a new sequence ° °
= inf ° f (s, ω )ds − g(s, ω )ds°
{h1 , φ 1 , h2 , φ 2 , · · · }, and denoted as p
g∈S (G)
Z t
0 0

{ f i = { f i (t, ω ) : t ∈ I, ω ∈ Ω} : i ∈ N}, 6 t inf k f (s, ω ) − g(s, ω )k2 ds.

g∈S p (G) 0

then for any (t, ω ) ∈ I × Ω,

By Lemma 1.3.12 in Ref.16, we have
F(t, ω ) ⊂ cl{ f i (t, ω ) : i ∈ N}.
Z t
By the definition of selection, for each n ∈ N and inf k f (s, ω ) − g(s, ω )k2 ds
a.e. (t, ω ) ∈ I × Ω, hn (t, ω ) ∈ F(t, ω ). Thus for a.e. g∈S p (G) 0
Z t
(t, ω ) ∈ I × Ω, = inf k f (s, ω ) − yk2 ds
0 y∈G(s,ω )
cl{ f i (t, ω ) : i ∈ N} ⊂ F(t, ω ). Z t
6 dH2 (F(s, ω ), G(s, ω ))ds.
Therefore, for a.e. (t, ω ) ∈ I × Ω, 0

F(t, ω ) = cl{ f i (t, ω ) : i ∈ N}. Noticing that

In addition, for each j ∈ N and a.e. (t, ω ) ∈ I × Ω,

Z t
sup inf kx − yk2
x∈Lt (F)(ω ) y∈Lt (G)(ω )
φ j (s, ω )ds ∈ Lt (F)(ω ),
0 = sup inf kIt ( f )(ω ) − yk2 ,
f ∈S p (F) y∈Lt (G)(ω )
then
Z t
cl{ φ j (s, ω )ds : j ∈ N} ⊂ Lt (F)(ω ). we have
0

Therefore, for a.e. (t, ω ) ∈ I × Ω, sup inf kx − yk2

x∈Lt (F)(ω ) y∈Lt (G)(ω )
Z t Z t
Lt (F)(ω ) = cl{ f i (s, ω )ds : i ∈ N}. 6 t dH2 (F(s, ω ), G(s, ω ))ds.
0 0

The proof is completed.

By the definition of Hausdorff metric and symmetri-
Now we prove an inequality of set-valued Leges-
cal property, we have (1).
gue integrals.
Theorem 6. Let p > 1. Then, for any F, G ∈
Corollary 7. If F ∈ L 2 (Kk (Rd )) and F(t) is con-
L p (Kk (Rd )), we have
tinuous in t with respect to the Hausdorff metric dH ,
dH2 (Lt (F)(ω ), Lt (G)(ω )) then Lt (F) is continuous in t with respect to dH .
Z t Proof. Take G = I[0,s] F, we have the conclusion
6t dH2 (F(s, ω ), G(s, ω ))ds. (1) by (1) and the classical dominated convergence the-
0
orem.

Published by Atlantis Press

Copyright: the authors
88
 Aumann Type Lebesgue Integral

4. Note solution of set-valued stochastic differential inclu-

sion (2) and its properties(Ref.5, 7). Thus, we can
Now we come back to the Black-Scholes formula for know the movement trajectory of stock price, which
the price of a European call option, the stock price changes in certain range when we predict the drift
st at the time t is assumed to satisfy of stock fluctuates or takes in a given interval. It is
very useful for us to make decision in the process
dst = st (udt + vdBt )
and system risk control.
where s0 > 0, u is the drift of stock, v is the volatil-
ity of stock and Bt is a Brownian motion. When we 5. Conclusions and Acknowledgement
predict the drift of stock fluctuates or takes in the in-
terval [u1 , u2 ], u1 , u2 ∈ R, then the stock price st at In this paper, we firstly discuss two basic prob-
the time t satisfies the following set-valued stochas- lems. 1) We prove that the Aumann type set-valued
tic differential inclusion Lebesgue integral of a set-valued stochastic process
is a set-valued stochastic process under the con-
dst ∈ [u1 st , u2 st ]dt + vst dBt , (2)
dition that the set-valued stochastic process takes
or set-valued stochastic integral form none-empty compact subset of Rd . 2) We discuss
³ Z t Z t ´ whether the former definition of the Aumann type
st − s0 ∈ clL p ∈ (L) [u1 st , u2 st ]dt + vst dBt , set-valued Lebesgue integral of a set-valued stochas-
0 0 tic process with respect to the time t is well-defined
where the first integral is Aumann type set-valued for all ω ∈ Ω. In this paper, we assume that A is µ -
Lebesgue integral, the second integral is classi- separable in probability space (Ω, A , µ ) so that the
cal Itb o integral. If st ∈ L p (R), then {Ft = former definition is available.
[u1 st , u2 st ], A t ,t ∈ I} ∈ L p (Kk (R)). If h = {ht ,t ∈ We secondly obtain the representation theorem,
I} ∈ L (R) satisfying u1 st (ω ) 6 ht (ω ) 6 u2 st (ω ) for and prove an important inequality of the Aumann
a.e. (t, ω ) ∈ I × Ω, then h ∈ S p (F) and S p (F) = type set-valued Lebesgue integrals of set-valued
{h = {ht ,t ∈ I} ∈ L (R) : u1 st (ω ) 6 ht (ω ) 6 stochastic processes with respect to t. These results
u2 st (ω ), for a.e.(t, ω ) ∈ I × Ω}. From Theorem are important for developing the theory of set-valued
2 and Theorem 3, we have SL1t (F) = {It (h) : h ∈ stochastic differential inclusions, which are useful in
finance and optimal control areas.
S p (F)}. By representation theorem, there exists a
Finally we would like to thank referees for
sequence of real-valued stochastic processes {hi =
their valuable suggestions and remarks, which im-
{hti ,t ∈ I}, i ∈ N} ⊂ S p (F) such that for a.e. (t0 , ω ) ∈
prove the presentation of this paper and thank
I × Ω,
M. Kisielewicz in Zielona Góra University and
F(t0 , ω ) = cl{hi (t0 , ω ) : i > 1}, Yukio Ogura in Saga University for the valuable
communication. This research is supported by
and NSFC(NO.10771010), Research fund of Beijing
Educational Committee, PHR(IHLB) and 111 Tal-
Lt0 (F)
n Z t0 o ent Project Fund of BJUT, P.R.China.
= cl hi (t, ω )dt : i > 1
0 1. M. Kisielewicz, “Set-valued stochastic integrals and
h Z t0 Z t0 i stochastic inclusions,” Stoch. Anal. Appl. 15, 783-800
= u1 st (ω )dt, u2 st (ω )dt (1997).
0 0 2. H. Richter, “Verallgemeinerung eines in der Statis-
Z t0
= [u1 , u2 ] st (ω )dt. tik benötigten Satzes der Masstherie,” Math. Annalen,
0 150, 85-90 (1963).
3. H. Kudo, “Dependent experiments and sufficient
Furthermore, we have the important inequality of statistics,” Nat. Sci. Rept. Ochanomizu Univ., Tokyo,
the integral by Theorem 6. So we can study the 4, 151-163 (1954).

Published by Atlantis Press

Copyright: the authors
89
 J. Li, S. Li

4. R. Aumann, “Integrals of set-valued functions,” J. Jour. Multivar. Anal., 7, 149-182 (1977).

Math. Anal. Appl., 12, 1-12 (1965). 16. S. Li, Y. Ogura and V. Kreinovich, Limit Theo-
5. M. Kisielewicz, “Set-valued stochastic integrals and rems and Applications of Set-Valued and Fuzzy Set-
stochastic inclusions,” Discuss. Math., 13, 119-126 Valued Random Variables, Kluwer Academic Publish-
(1993). ers, 2002.
6. M. Kisielewicz, Differential Inclusions and Optimal 17. J.A. Yan, Measure Theorey, Science Press, 2004.
Control, Kluwer Academic Publishers, New York, 18. I. Molchanov, Theory of Random Sets, Springer, 2005.
1991. 19. B.K. Kim and J.H. Kim, “Stochastic integrals of set-
7. M. Kisielewicz, “Properties of solution set of stochas- valued processes and fuzzy processes,” J. Math. Anal.
tic inclusions,” J. Appl. Math. Stoch. Anal., 6, 217-236 Appl., 236, 480-502 (1999).
(1993). 20. J.P. Aubin and H. Frankowska, Set-Valued Analysis,
8. M. Kisielewicz, M. Michta and J. Motyl, “Set-valued Birkhauser, 1990.
approach to stochastic control part I (existence and 21. B.D. Craven, Lebesgue Measure and Integral, Pitman,
regularity properties),” Dynamic Syst. and Appl., 12, 1982.
405-432 (2003). 22. C. Castaing and M. Valadier, Convex Analysis and
9. M. Kisielewicz, M. Michta and J. Motyl, “Set-valued Measurable Multifunctions, Lect. Notes in Math., 580,
approach to stochastic control part II (viability and Springer–Verlag, Berlin, New York, 1977.
semimartingale issues),” Dynamic Syst. and Appl., 12, 23. S. Hu and N.S. Papageorgiou, Handbook of Multival-
433-466 (2003). ued Analysis, Kluwer Academic Publishers, 1997.
10. J. Li and S. Li, “Set-valued stochastic Lebesgue inte- 24. E.J. Jung and J.H. Kim, “On set-valued stochastic in-
gral and representation theorems,” International Jour- tegrals,” Stoch. Anal. Appl., 21, 401-418 (2003).
nal of Computational Intelligence Systems, 1, 177-187 25. E. Klein and A.C. Thompson, Theory of Correspon-
(2008). dences Including Applications to Mathematical Eco-
11. G. Da Prato and H. Frankowska,“ A stochastic Filip- nomics, John Wiley & Sons, 1984.
pov theorem,” Stoch. Anal. Appl., 12, 409-426 (1994). 26. S. Li, Y. Ogura, F.N. Proske and M.L. Puri, “Central
12. J. Motyl, “Existence of solutions of set-valued Ito limit theorems for generalized set-valued random vari-
equation,” Bull. Acad. Pol. Sci., 46, 419-430 (1998). ables,” J. Math. Anal. Appl., 285, 250-263 (2003).
13. Y. Ogura, On stochastic differential equations with set 27. S. Li and A. Ren, “Representation theorems, set-
coefficients and the Black-Scholes model, in the Pro- valued and fuzzy set-valued Ito intergal,” Fuzzy Sets
ceeding of the 8th International Conference on Intel- and Syst., 158, 949-962 (2007).
ligent Technologies, in Sydney, 300-304, 2007. 28. N.S. Papageorgiou, “On the conditional expectation
14. W. Zhang, S. Li, Z. Wang and Y. Gao, An Intro- and convergence properties of random sets,” Trans.
duction of Set-Valued Stochastic Processes, Science Amer. Math. Soc., 347, 2495-2515 (1995).
Press, Beijing/New York, 2007. 29. J. Zhang, S. Li, I. Mitoma and Y. Okazaki, “On set-
15. F. Hiai and H. Umegaki, “Integrals, conditional ex- valued stochastic integrals in an M-type 2 Banach
pectations and martingales of multivalued functions,” space,” J. Math. Anal. Appl., 350, 216-233 (2009).

Published by Atlantis Press

Copyright: the authors
90