Statistics and Probability Letters 78 (2008) 28642867

Contents lists available at ScienceDirect

Statistics and Probability Letters

journal homepage: www.elsevier.com/locate/stapro

An existence theorem for stochastic functional differential equations

with delays under weak assumptions
Nikolaos Halidias a , Yong Ren b,
a Department of Statistics and Actuarial-Financial Mathematics, University of the Aegean, Karlovassi 83200 Samos, Greece
b Department of Mathematics, Anhui Normal University, Wuhu 241000, PR China

article

a b s t r a c t

info

Article history:
Received 21 March 2008
Received in revised form 10 April 2008
Accepted 11 April 2008
Available online 22 April 2008

In this paper, we consider stochastic functional differential equations with delays and our
aim is to prove an existence theorem when the drift coefficient is not continuous.
2008 Elsevier B.V. All rights reserved.

1. Introduction
In this paper we consider the following stochastic functional differential equations with finite delay (SFDEDs in short)
dX (t) = f (t, Xt ) dt + g(t, Xt ) dB(t),

t0 t T,

(1)

where X (t) denotes the value of the stochastic process at time t and Xt = {X (t + ) : 0} could be considered
as a C ([, 0]; Rd )-valued stochastic process. Recall that with C ([, 0]; Rd ) we denote the family of bounded continuous
Rd -valued functions defined on [, 0] with norm kk = sup0 |()|. We define the initial value of (1) as follows:
Xt0 =

= {() : 0} is Ft0 -measurable,

C ([, 0]; Rd )-valued random variable

such that M 2 ([, 0]; Rd ),

(2)

where M ([, 0]; R ) denote the family of all Ft0 -measurable, R -valued process (t) = (t, ), t [, 0] such that
R0
E |(t)|2 dt < .
For system (1) with initial value (2), we can write it in the following integral form:
Z t
Z t
X (t) = (0) +
f (s, Xs )ds +
g(s, Xs )dB(s), t0 t T.
2

t0

t0

Let f : [t0 , T ] C ([, 0]; R ) Rd and g : [t0 , T ] C ([, 0]; Rd ) Rdm be both Borel measurable.
Let us state the following conditions. For any , C ([, 0]; Rd ) and t [t0 , T ], suppose that:
d

(H1) |f (t, ) f (t, )|2 K k k2 , K > 0.

For any (t, ) [t0 , T ] C ([, 0]; Rd ), suppose that
(H2) |f (t, )|2 K (1 + kk2 ), K > 0.
Condition (H1) is the classical Lipschitz assumption and (H2) is the linear growth condition.

Corresponding author.
E-mail addresses: nick@aegean.gr (N. Halidias), brightry@hotmail.com (Y. Ren).
0167-7152/$ see front matter 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.spl.2008.04.006

N. Halidias, Y. Ren / Statistics and Probability Letters 78 (2008) 28642867

2865

When f , g satisfies conditions (H1) and (H2) then it is well known that our problem has a unique solution X (t)
M2 ([t0 , T ]; Rd ), see for example Mao (1997). In this paper we are interested in the existence part of our problem. Using
some simple ideas we will show an existence theorem when f is increasing but not continuous, such as in the following
SFDEDs:
dX (t) = H(Xt )dt + dB(t),
where H : R R is the Heaviside function, which is defined by

0, x < 0,
H(x) :=
1, x 0.
Such equations arise, for example, when one considers the effects of background noise switching systems with delays. Some
ideas have borrowed from Halidias and Kloeden (2006) (see also the references therein), where strong existence theorems
with discontinuous drift coefficient for stochastic differential equations has been proved. To the best of our knowledge there,
does not exist any such existence result for SFDEDs. For a numerical solution in this case see Halidias and Kloeden (2008).
For some existence and uniqueness results on stochastic functional differential equations with infinite delay please see Ren
et al. (2007) and the references therein. Finally, for some applications of SFDEDs on mathematical finance see Stoica (2004).
As the first step, we only pay our attention to deriving the existence of solutions to SFDEDs under the condition of the drift
coefficient being not continuous by a simple but interesting Comparison Theorem. We will do our best to use the results
obtained in this paper for mathematical finance in the further study.
2. Notations and auxiliary results
Let | | denote the Euclidean norm in Rd . If p
A is a vector or a matrix, its transpose is denoted by AT ; if A is a matrix,
its Frobenius norm is represented by |A| =
trace(AT A). Let t0 be a positive constant and ( , F , P) be a complete
probability space with a filtration {Ft }tt0 satisfying the usual conditions. Assume that B(t) is an mdimensional Brownian
motion defined on ( , F , P), that is B(t) = (B1 (t), B2 (t), , Bm (t))T . For a stochastic process X (t), t0 t T , let
X = supt0 tT |X (t)|. Let S2 denote the space of all Ft -adapted process X (t), t0 t T such that kX kS2 = kX kL2 < ,
with L2 = L2 ( , Rd ). It follows that (S2 , k kS2 ) is a Banach space.
With all the above preparation, we give the definition of solutions of Eq. (1) with initial value (2).
Definition 1. Rd -valued stochastic process X (t) defined on t T is called the solution of (1) with initial value (2), if:
(i) X (t) is pathwise continuous and for all t0 t T, X (t) is Ft adapted;
(ii) {f (t, Xt )} L1 ([t0 , T ]; Rd ) and {g(t, Xt )} LR2 ([t0 , T ]; Rdm )R;
t
t
(iii) Xt0 = , for each t0 t T, X (t) = (0) + t0 f (s, Xs )ds + t0 g(s, Xs )dB(s), a.s.
Our first result is a comparison theorem for SFDEDs. Consider the following problems,
Z t
Z t
X (t) = 1 (0) +
f1 (s, Xs )ds +
g(s, Xs )dB(s), t0 t T,
t0

X (t ) =

2 (0) +

t0

(3)

t0

f2 (s, Xs )ds +

t0

g(s, Xs )dB(s),

t0 t T.

(4)

Suppose that 1 2 a.s. componentwise and f1 (t, x) f2 (t, x) componentwise for all t [t0 , T ], x C ([, 0]; Rd ). We
want to prove the following theorem.
Theorem 2. Assume that f1 or f2 satisfies the Lipschitz condition (H1) and that f1 , f2 , g satisfy the linear condition (H2). Then, if
X 1 is a continuous solution of problem (3) and X 2 of problem (4) then X 1 X 2 for all t a.s.
Proof. Suppose that f2 satisfies the Lipschitz condition. We define the truncation and penalization functions as follows
p(x, y) = max{x, y},

r(x) =

p(x) x

1 + | x |2

It is easy to see that |p(x, y) p(x, z)| |y z| and the same holds for r(). We consider now the following equation
Z t
Z t
g(s, p(X 1 , Xs ))dB(s), t0 t T.
X (t) = 2 (0) +
f2 (s, p(Xs1 , Xs ) + r(Xs ))ds +
t0

(5)

t0

We can see that the coefficients satisfy conditions (H1), (H2) and so, (5) has a unique solution, say Y (t).
Suppose that there exists an interval (t1 , t2 ) [t0 , T ] such that Y (t1 ) = X 1 (t1 ) and X 1 (t) Y (t) for all t (t1 , t2 ). Then, it
follows that
Z t2
Z t2
Z t2
Y (t) X 1 (t) =
(f2 (s, p(Xs1 , Ys )) f1 (s, Xs1 ))ds +
(g(s, p(X 1 , Ys )) g(s, Xs1 ))dB(s) +
r(Ys )ds 0.
t1

t1

t1

N. Halidias, Y. Ren / Statistics and Probability Letters 78 (2008) 28642867

2866

From the definition of p, r we have p(Xt1 , Yt ) = Xt1 in (t1 , t2 ), and r(Yt ) 0, thus it follows that Y (t) X 1 (t) for all
t (t1 , t2 ), which is a contradiction. Thus, Y X 1 a.s. and so p(X 1 , Y ) = Y, r(Y ) = 0 and this means that in fact Y = X 2 X 1
because (5) has a unique solution X 2 .
The proof remains in fact the same, with minor modifications, if f1 satisfies the Lipschitz condition.
Note also, that if we have strict inequality f1 < f2 then we can construct a Lipschitz function between them. So, in this
case we do not need either f1 or f2 to be Lipschitz continuous. 
3. SFDEs with discontinuous drift coefficient
In this section, we assume that f is increasing, left continuous and such that f (t, x) 0 for all (t, x) [t0 , T ] C ([, 0]; Rd )
but not continuous. We construct the following sequence of problems.
X n (t) =

(0) +

t0

f (s, Xsn1 )ds +

t0

g(s, Xsn )dB(s),

t0 t T,

(6)

where X 0 = Lt with Lt is the unique solution of the following problem,

Lt =

t0

g(s, Ls )dBs .

(7)

So, X 1 Lt using the fact that f 0 and the comparison theorem. Thus, we see that X n is an increasing sequence and we
shall prove that X n is bounded in S2 norm.
Lemma 3. If X n is a solution of problem (6) then,
!
E

sup

t0 sT

|X n (s)|2 C

(8)

where C is a positive constant independent of n.

Proof. We are going to use the same arguments as in Mao (1997), Lemma 5.2.3. So, for every n 1 we define the stopping
time

m = T inf {t [t0 , T ] : kXtn k m}.

Note that m T and define X n,m (t) = X n (t m ) for t [t0 , T ]. Then for t [t0 , T ],
Z t
Z t
X n,m (t) = (0) +
f (s, Xsn,m )I[t0 ,m ] (s)ds +
g(s, Xsn,m )I[t0 ,m ] (s)dBs .
t0

t0

Using the linear growth condition (H1) and the Holder inequality we have the following,
Z t
Z t
E|X n,m (t)|2 3Ekk2 + C1
K (1 + kXsn1,m k2 ) + C2
K (1 + kXsn,m k2 )ds,
t0

t0

Then, for any k N we have that

max E|X n,m (t)|2 3Ekk2 + C1

1nk

Z t
t0



Z t
1 + max kXsn1,m k2 ds + C2
1 + max kXsn,m k2 ds.
1nk

1nk

t0

Thus, using Doobs martingale inequality, for any n, m N we have that

!
Z
E

sup |X n,m (s)|2

t0 st

3Ekk2 + C

t0

(1 + EkXsn,m k2 )ds.

On the other hand as in Mao (1997), Lemma 5.2.3 we can obtain

!
Z
1+E

sup |X n,m (s)|2

t0 st

1 + 4Ekk2 + C

t0

1 + E sup |X n,m (r)|2 ds.

t0 rs

Now, using the Gronwall inequality and taking the limit as m we get the desired result.

Theorem 4. Let f be increasing in the second variable, left continuous and such that f (t, x) 0 for all (t, x) [t0 , T ]
C ([, 0]; Rd ). Then there exists at least one solution X M 2 ([t0 , T ], Rd ).

N. Halidias, Y. Ren / Statistics and Probability Letters 78 (2008) 28642867

2867

Proof. From Theorem 2, (X n ) is increasing, and Lemma 3 shows that it is bounded in S2 . We deduce from the Dominated
Convergence Theorem that X n converges in S2 . We denote by X the limit of (X n ). So we get for almost all ,
f (t, X n (t)) f (t, X (t)),

dt a.s., n .

(9)

and

|f (t, X n (t))| K (1 + sup |X n (t)|) L1 ([t0 , T ], dt).

(10)

Thus, for almost all and uniformly in t,

Z t
Z t
f (s, X n (s))ds
f (s, X (s))ds,
t0

t0

n .

From the property of g and the continuity property of the stochastic integral, we get
Z t

Z t

 P
sup  g(s, X n (s)dB(s))
g(s, X (s)dB(s)) 0, n .
t0 tT

t0

(11)

(12)

t0

Noting that X n is an increasing sequence we have also that the above convergence is also P-a.s. Finally, it is easy to prove
that X n X converges uniformly in t, so X is also continuous. 
So, taking limits in (6), we deduce that X is our desired solution for (1) with initial value (2).
Acknowledgements
The authors wish to thank the anonymous referee for his or her valuable comments, correcting errors and improving
written language. This work was supported by The National Natural Science Foundation of China (no. 10726075).
References
Halidias, N., Kloeden, P., 2006. A note on strong solutions for stochastic differential equations with discontinuous drift coefficient. J. Appl. Math. Stoch. Anal.
16. doi:10.1155/JAMSA/2006/73257. Article ID 73257.
Halidias, N., Kloeden, P., 2008. A note on the EulerMaruyama scheme for stochastic differential equations with a discontinuous monotone drift coefficient.
BIT 48, 5159.
Mao, X., 1997. Stochastic Differential Equations and Applications. In: Horwood Series in Mathematics and Applications, Chichester.
Ren, Y., Lu, S., Xia, N., 2007. Remarks on the existence and uniqueness of the solutions of stochastic functional differential equations with infinite delay. J.
Comput. Appl. Math. doi:10.1016/j.cam.2007.08.002.
Stoica, G., 2004. A Stochastic delay financial model. Proc. Amer. Math. Soc. 133, 18371841.