Applied Mathematics and Computation 215 (2010) 4008–4021

Contents lists available at ScienceDirect

Applied Mathematics and Computation

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

Convergence and stability of numerical solutions to a class of index 1

stochastic differential algebraic equations with time delay
Xiaomei Qu a,*, Chengming Huang a, Chao Liu b
a
School of Mathematics and Statistics, Huazhong University of Science and Technology, Wuhan 430074, PR China
b
College of Computer Science and Technology, Huanggang Normal University, Huangzhou 438000, PR China

a r t i c l e i n f o a b s t r a c t

Keywords: In this paper, we study the convergence and stability of the stochastic theta method (STM)
Stochastic delay differential algebraic for a class of index 1 stochastic delay differential algebraic equations. First, in the case of
equation constrained mesh, i.e., the stepsize is a submultiple of the delay, it is proved that the
Convergence method is strongly consistent and convergent with order 1/2 in the mean-square sense.
Mean-square
Then, the result is further extended to the case of non-constrained mesh where we employ
Stability
Stochastic theta method
linear interpolation to approximate the delay argument. Later, under a sufficient condition
for mean-square stability of the analytical solution, it is proved that, when the stepsizes are
sufficiently small, the STM approximations reproduce the stability of the analytical solu-
tion. Finally, some numerical experiments are presented to illustrate the theoretical
findings.
Crown Copyright Ó 2009 Published by Elsevier Inc. All rights reserved.

1. Introduction

In recent years, because stochastic differential equations (SDEs) play an important role in modeling of problems in many
branches of science and industry, their numerical analysis has received considerable attention; see, e.g., [9–12,14,15]. During
the same period of time, much work has also been done in the field of numerical solution of differential algebraic equations
(DAEs) (see [1,7,16,18]). These systems can be found in a wide variety of scientific and engineering applications, including
circuit analysis, computer-aided design and real-time simulation of mechanical systems, and optimal control. While delay
differential equations (DDEs) serve as models of physical processes whose time evolution depends on their past history,
which arise from, for example, circuit simulation and power systems. The same as SDEs and DAEs, DDEs are also well known
and intensively studied (cf. [2,5,19]).
A generalization of SDEs and DAEs is stochastic differential algebraic equations (SDAEs). Not much work has been done on
numerical methods for SDAEs. In [20], Winkler gives a introduction to index 1 SDAEs and discusses the convergence and sta-
bility of their numerical methods.
Delay differential algebraic equations (DDAEs), which can be seen as an extension of DAEs and DDEs, appear frequently in
many problems [22]. Until now, there is little work on numerical analysis of DDAEs, and most of papers are concerned with
linear problems. Here, we refer to [6] for a recent discussion of numerical stability for linear DDAEs.
Stochastic delay differential algebraic equations (SDDAEs), which have both delay and algebraic constraints, possess sto-
chastic nature as well. Relatively little is known about the numerical analysis of SDDAEs. In this paper, we consider a class of
index 1 SDDAEs. Their stochastic theta method is established, and the convergence and mean-square stability are analyzed.
This paper is organized as follows.

* Corresponding author.
E-mail addresses: maggiequ111@163.com (X. Qu), chengming_huang@hotmail.com (C. Huang), leochal@163.com (C. Liu).

0096-3003/$ - see front matter Crown Copyright Ó 2009 Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2009.12.007
 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021 4009

In Section 2, a class of index 1 SDDAEs is given and the initial value problems are formulated. We introduce some nec-
essary notations and assumptions, then we present the conditions of existence and uniqueness of strong solutions, and
establish the stochastic theta method for index 1 SDDAEs.
In Section 3, we provide a proof of mean-square consistency and convergence for the stochastic theta method applied to
index 1 SDDAEs. This generalizes some results of strong convergence for index 1 SDAEs in [20] to index 1 SDAEs with time
delay. We introduce an attempt to treat the delay argument by linear interpolation, and also prove that the stochastic theta
method for SDDAEs with linear interpolation procedure is consistent and convergent.
In Section 4, we first discuss the stability of analytical solution for a class of index 1 SDDAEs. Then we prove that the STM
approximate solutions can preserve MS-stability of the analytical solution, when the stepsizes are sufficiently small.
Finally, in Section 5 some numerical experiments are presented to illustrate the theoretical findings.

2. Index 1 SDDAEs and the stochastic theta method

Throughout this paper, we let ðX; F; fFt gtPt0 ; PÞ be a complete probability space with a filtration fFt gtPt0 satisfying the
usual conditions (i.e., it is increasing and right continuous while F0 contains all P-null sets). Let W ¼ ðW 1 ; W 2 ; . . . ; W s ÞT be a
s-dimensional Brownian motion defined on the probability space. Let j  j be the Euclidean norm in Rd . If A is a vector or ma-
trix, its transpose is denoted by AT . If A is a matrix, its trace norm is denoted by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jAj ¼ traceðAT AÞ:

Lets > 0; J ¼ ½t0 ; T and Rþ ¼ ½0; 1Þ. Denote by Cð½s; 0; Rd Þ the family of continuous functions from ½s; 0 to Rd with the
norm kuk ¼ sups6h60 juðhÞj. Let p > 0, and denote by LpFt ðJ; Rd Þ the family of Ft0  measurable CðJ; Rd Þ-valued random vari-
0

ables n ¼ fnðhÞ : h 2 Jg such that Eknkp < 1. The norm and the inner product on L2 ðX; F; PÞ are defined by
1
kxkL2 ¼ ðEjxj2 Þ2 ; hx; yi ¼ EðxT yÞ:

2.1. Index 1 SDDAEs

In the following, we consider a class of index 1 stochastic delay differential algebraic equations (SDDAEs) in Itô sense

dxðtÞ ¼ f ðxðtÞ; xðt  sÞ; yðtÞÞdt þ gðxðtÞ; xðt  sÞ; yðtÞÞdWðtÞ; t 2 J;
ð2:1Þ
0 ¼ uðxðtÞ; xðt  sÞ; yðtÞÞ; t 2 J;
with initial data xðtÞ ¼ uðtÞ; yðtÞ ¼ wðtÞ; t0  s 6 t 6 t 0  xðtÞ and yðtÞ are Rd and Rl -valued processes, respectively,
f : Rd  Rd  Rl ! Rd , u : Rd  Rd  Rl ! Rl , g : Rd  Rd  Rl ! Rds are given continuous functions, uðtÞ and wðtÞ are Ft0 -mea-
surable and
!
2 2
Ekuk < 1; Ekwk < 1 kuk ¼ sup juðtÞj :
t 0 s6t6t0

For uðx; m; yÞ ¼ 0, the Jacobian matrix @u=@y is supposed to have a uniformly bounded inverse. Without loss of generality, we
suppose

jð@uðx; v ; yÞ=@yÞ1 j 6 L1 ; j@uðx; v ; yÞ=@xj 6 L2 ; j@uðx; v ; yÞ=@ v j 6 L3 : ð2:2Þ

Integrating on both sides of (2.1) we can obtain a stochastic integral problem

8 Rt Rt
>
> xðtÞ ¼ uðt 0 Þ þ t0 f ðxðsÞ; xðs  sÞ; yðsÞÞds þ t0 gðxðsÞ; xðs  sÞ; yðsÞÞdWðsÞ; t 2 J;
<
0 ¼ uðxðtÞ; xðt  sÞ; yðtÞÞ; ð2:3Þ
>
>
:
xðtÞ ¼ uðtÞ; yðtÞ ¼ wðtÞ; t 0  s 6 t 6 t0 ;

where the second integral in the first equation of (2.3) is an Itô integral.

2.2. Existence and uniqueness of the solution

In order to ensure the existence and uniqueness of the solution, we further assume that f and g are smooth enough and
satisfy
ðiÞ jf ðx; v ; yÞ  f ðx; v ; y
Þj _ jgðx; v ; yÞ  gðx; v ; y
Þj 6 b1 ðjx  xj þ jv  v j þ jy  y
jÞ; ð2:4Þ
ðiiÞ jf ðx; v ; yÞj _ jgðx; v ; yÞj 6 b2 ð1 þ jxj þ jv j þ jyjÞ; ð2:5Þ
ðiiiÞ uðuðtÞ; uðt  sÞ; wðtÞÞ ¼ 0; t0  s 6 t 6 t 0 ; ð2:6Þ
4010 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021

for all x; v ;   ; 2 Rd ; y; y
x; v  2 Rl , where b1 ; b2 > 0 are constants (Generally, (i) is called global Lipschitz condition (GLC), (ii) is
known as linear growth condition (LGC)). Because the Jacobian matrix @u=@y is supposed to have a uniformly bounded in-
verse, so using the existence theorem of implicit function, then for uðx; m; yÞ ¼ 0 there exists a unique solution

y ¼ Uðx; v Þ;

which implies

yðtÞ ¼ UðxðtÞ; xðt  sÞÞ: ð2:7Þ

Substituting (2.7) into (2.1), we can get

dxðtÞ ¼ f ðxðtÞ; xðt  sÞ; UðxðtÞ; xðt  sÞÞÞdt þ gðxðtÞ; xðt  sÞ; UðxðtÞ; xðt  sÞÞÞdWðtÞ; t 2 J: ð2:8Þ

Thus, the index 1 SDDAE (2.1) can be transformed to a stochastic delay differential equation theoretically. Nevertheless, it is
normally difficult to find the explicit expression of Uðx; mÞ so that we have to solve the original SDDAE.
Lemma 2.1 [12]. For stochastic delay differential equations
(
dxðtÞ ¼ ~f ðt; xðtÞ; xðt  sÞÞdt þ g~ðt; xðtÞ; xðt  sÞÞdWðtÞ; t > t0 ;
ð2:9Þ
xðtÞ ¼ n; t 6 t0 ;

suppose f and g are globally Lipschitz-continuous with respect to x, continuous with respect to t, satisfy the linear growth condition
(LGC), and that initial data Eknk2 < 1, which is independent of the Brownian motion W and with finite second moments. Then
there exists a solution process xðÞ of SDDEs (2.9) that is pathwise unique. Moreover, the solution process xðÞ is square-integrable
and fulfills
!
E sup jxðsÞj2 6 c1 Eð1 þ knk2 Þec2 ðtt0 Þ ;
t 0 s6s6t

with constants c1 ; c2 .

The regular SDDE (2.9) together with the assembling of the solution (2.7) is equivalent to the problem (2.8). Based on this
fact we are now able to give our theorem on the existence and uniqueness of solution to index 1 SDDAEs.
Theorem 2.1. Suppose f and g satisfy (i)–(iii), and that uðtÞ 2 L2Ft ðJ; Rd Þ; wðtÞ 2 L2Ft ðJ; Rl Þ are independent of the Brownian
0 0
motion W and with finite second moments. Then there exists a solution process xðÞ of index 1 SDDAEs (2.1) that is pathwise unique.
Moreover, the solution process xðtÞ is square-integrable.
The proof of this theorem is analogue to the theorem on existence and uniqueness of solution to neutral stochastic func-
tional differential equations with infinite delay in [17]. Hence, throughout this paper, we assume that (2.1) has a unique glo-
bal solution ðxðtÞ; yðtÞÞ.

2.3. Stochastic theta method for index 1 SDDAEs

Apply stochastic theta method to (2.1), then we have

8
>
> x ¼ xn þ h½hf ðxnþ1 ; xnþ1 ; ynþ1 Þ þ ð1  hÞf ðxn ; xn ; yn Þ þ gðxn ; xn ; yn ÞMW n ; n > 0;
< nþ1
0 ¼ uðxnþ1 ; xnþ1 ; ynþ1 Þ; ð2:10Þ
>
>
:
xn ¼ uðtn Þ; yn ¼ wðt n Þ; n 6 0;

where the stepsize h > 0, the parameter h 2 ½0; 1; tn ¼ nh; n ¼ 0; 1; . . . N, MW n ¼ Wðtnþ1 Þ  Wðtn Þ, each xn is an approximation
to xðtn Þ. For h ¼ 0, (2.10) reduces to Euler–Maruyama (EM). The h ¼ 1 case is addressed as backward Euler (BE).
The argument  xn denotes an approximation to xðtn  sÞ, that is obtained by an interpolation procedure at the point
t ¼ tn  s using values fxk gk6n .
Let s ¼ ðm  dÞh with m a positive integer and d 2 ½0; 1Þ. We can define

xk ¼ dxkmþ1 þ ð1  dÞxkm ; ð2:11Þ

here when k 6 0; xk ¼ uðt0 þ khÞ; yk ¼ wðt 0 þ khÞ.

Integrating from t n to tnþ1 on (2.1), we have
Z t nþ1 Z tnþ1
xðt nþ1 Þ ¼ xðt n Þ þ f ðxðtÞ; xðt  sÞ; yðtÞÞdt þ gðxðtÞ; xðt  sÞ; yðtÞÞdWðtÞ; t 2 ½t n ; tnþ1 : ð2:12Þ
tn tn
 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021 4011

3. Convergence analysis

Throughout this section, we always assume that conditions (i)–(iii) are satisfied. For the convenience of discussion, we
denote

r n :¼ xðtnþ1 Þ  xðt n Þ  h½hf ðxðtnþ1 Þ; xðtnþ1  sÞ; yðtnþ1 ÞÞ
ð3:1Þ
þð1  hÞf ðxðt n Þ; xðtn  sÞ; yðt n ÞÞ  gðxðtn Þ; xðt n  sÞ; yðt n ÞÞMW n ;
and Et ðxÞ ¼ EðxjFt Þ. In accordance with [4], we present the following definitions.
Defintion 3.1. The numerical method (2.10) is strongly consistent with order p in the mean-square sense if the following
estimates hold
pþ1 pþ1=2
max kEtn ðr n ÞkL2 6 Ch ; max krn kL2 6 Ch ;
06n6N 06n6N

where C is a positive constant independent of h.

Defintion 3.2. The numerical method (2.10) is strongly convergent with order p in the mean-square sense if the following
estimate holds
p
max ðkxn  xðt n ÞkL2 þ kyn  yðtn ÞkL2 Þ 6 Ch ;
06n6N

where C is a positive constant independent of h.

In order to prove our main results, we need to introduce a lemma, whose proof can be found in [12].
Lemma 3.1. If ðxðtÞ; yðtÞÞ is the analytical solution of (2.1), and conditions (i)–(iii) hold, then
!
E sup jxðtÞj2 6 c1 ; ð3:2Þ
t 0 s6t6T

and for all t0  s 6 s < t 6 T, the estimate holds

EjxðtÞ  xðsÞj2 6 c2 ðt  sÞ; ð3:3Þ

where c1 ; c2 are constants independent of s and t.

Theorem 3.1. The stochastic theta method (2.10) applied to index 1 SDDAEs (2.1) is strongly consistent with order 12.

Proof. By the properties of Itô integral (see [12]), we have

Z t nþ1 
Etn ½gðxðtÞ; xðt  sÞ; yðtÞÞ  gðxðt n Þ; xðtn  sÞ; yðt n ÞÞdWðtÞ ¼ 0:
tn

Substituting (2.12) into (3.1), we have

Z tnþ1 Z t nþ1
rn ¼ f ðxðtÞ; xðt  sÞ; yðtÞÞdt  ½hf ðxðtnþ1 Þ; xðtnþ1  sÞ; yðtnþ1 ÞÞ þ ð1  hÞf ðxðt n Þ; xðt n  sÞ; yðtn ÞÞdt
tn tn
Z tnþ1
þ ½gðxðtÞ; xðt  sÞ; yðtÞÞ  gðxðt n Þ; xðtn  sÞ; yðt n ÞÞdWðtÞ: ð3:4Þ
tn

By the inequality ðEXÞ2 6 EX 2 , we arrive at

kEtn ðr n Þk2L2 ¼ EðjEtn ðr n Þj2 Þ

 Z tnþ1 2 !
 

¼ E Etn ½ ðf ðxðtÞ; xðt  sÞ; yðtÞÞ  hf ðxðt nþ1 Þ; xðt nþ1  sÞ; yðt nþ1 ÞÞ  ð1  hÞf ðxðtn Þ; xðtn  sÞ; yðt n ÞÞÞdt
tn
  Z
 t nþ1
6 E Etn h ½f ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðtnþ1 Þ; xðtnþ1  sÞ; yðt nþ1 ÞÞdt
tn
Z 2 !
tnþ1 
þ ð1  hÞ ½f ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðt n Þ; xðt n  sÞ; yðt n ÞÞdt  : ð3:5Þ
tn

2 Rb pffiffiffiffiffiffiffiffiffiffiffiffiffiffiR b 2 12
By the inequality ðha þ ð1  hÞbÞ2 6 ha2 þ ð1  hÞb ; h 2 ½0; 1 and Hölder inequality a
1  jf jdt 6 jb  aj a jf j dt , we can
see from (3.5) that
4012 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021

" Z 2
t nþ1
kEtn ðr n Þk2L2 6 E hEtn jf ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðt nþ1 Þ; xðt nþ1  sÞ; yðt nþ1 ÞÞjdt
tn

Z !2 3
t nþ1
þ ð1  hÞEtn jf ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðt n Þ; xðt n  sÞ; yðtn ÞÞjdt 5
tn
Z t nþ1
6 hE hEtn jf ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðtnþ1 Þ; xðtnþ1  sÞ; yðtnþ1 ÞÞj2 dt
tn
Z tnþ1 #
2
þð1  hÞEtn jf ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðtn Þ; xðt n  sÞ; yðt n ÞÞj dt : ð3:6Þ
tn
R tnþ1
We set Ri ¼ tn
jf ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðti Þ; xðti  sÞ; yðt i ÞÞj2 dt, for i ¼ n; n þ 1. By GLC (2.4), we derive that
Z t nþ1
Ri 6 b21 ½jxðtÞ  xðt i Þj þ jxðt  sÞ  xðt i  sÞj þ jyðtÞ  yðt i Þj2 dt: ð3:7Þ
tn

Because the Jacobian matrix @u=@y is supposed to have a globally bounded inverse, so using the existence theorem of implicit
function and (2.2), one has
  !
 @uðx; v ; yÞ1 @uðx; v ; yÞ 
 
jU x ðx; v Þj ¼   6 L1 L2 ; ð3:8Þ
 @y @x 
   !
 @uðx; v ; yÞ 1 @uðx; v ; yÞ 
 
jU v ðx; v Þj ¼   6 L1 L3 : ð3:9Þ
 @y @v 
By (3.8) and (3.9), we may ensure that
jyðtÞ  yðtn Þj 6 L1 L2 jxðtÞ  xðtn Þj þ L1 L3 jxðt  sÞ  xðt n  sÞj: ð3:10Þ
Taking conditional expectation on both sides of inequality (3.7) and using (3.10) and Lemma 3.1, we have
Z tnþ1
Etn jRi j 6 b21 E ½ð1 þ L1 L2 ÞjxðtÞ  xðti Þj þ ð1 þ L1 L3 Þjxðt  sÞ  xðt i  sÞj2 dt
tn
Z t nþ1 Z t nþ1

6 2b21 ð1 þ L1 L2 Þ2 EjxðtÞ  xðt i Þj2 dt þ ð1 þ L1 L3 Þ2 Ejxðt  sÞ  xðti  sÞj2 dt

tn tn
2
6 c2 b21 h ½ð1 þ L1 L2 Þ2 þ ð1 þ L1 L3 Þ2 ; ð3:11Þ
for i ¼ n; n þ 1. Substituting (3.11) into (3.6) gives
3 3
kEtn ðr n Þk2L2 6 c2 b21 ½ð1 þ L1 L2 Þ2 þ ð1 þ L1 L3 Þ2 h 6 c2 b21 ð2 þ L1 L2 þ L1 L3 Þ2 h ; ð3:12Þ
which implies
pffiffiffiffiffi 3
kEtn ðr n ÞkL2 6 c2 b1 ð2 þ L1 L2 þ L1 L3 Þh2 : ð3:13Þ
Moreover, because gðxðtÞ; xðt  sÞ; yðtÞÞ 2 L2 ðJ; Rds Þ, we have (see [12])
Z 2
 t nþ1 
E ½gðxðtÞ; xðt  sÞ; yðtÞÞ  gðxðt n Þ; xðtn  sÞ; yðtn ÞÞdwðtÞ
tn
Z t nþ1 
¼E jgðxðtÞ; xðt  sÞ; yðtÞÞ  gðxðt n Þ; xðtn  sÞ; yðtn ÞÞj2 dt :
tn
2
By ða þ bÞ2 6 2a2 þ 2b , we can compute

kr n k2L2 ¼ Ejr n j2
Z tnþ1
6 2hE hjf ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðt nþ1 Þ; xðtnþ1  sÞ; yðtnþ1 ÞÞj2 dt
tn
Z t nþ1

þ ð1  hÞ jf ðxðtÞ; xðt  sÞ; yðtÞÞ  f ðxðt n Þ; xðt n  sÞ; yðt n ÞÞj2 dt

tn
Z tnþ1
þ2 EjgðxðtÞ; xðt  sÞ; yðtÞÞ  gðxðt n Þ; xðt n  sÞ; yðtn ÞÞj2 dt
tn
Z t nþ1 h i
6 2h½hERnþ1 þ ð1  hÞERn  þ 4b21 ð1 þ L1 L2 Þ2 EjxðtÞ  xðt n Þj2 þ ð1 þ L1 L3 Þ2 Ejxðt  sÞ  xðt n  sÞj2 dt
tn
2
6 4b21 c2 ð2 þ L1 L2 þ L1 L3 Þ2 h ; ð3:14Þ
 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021 4013

hence
pffiffiffiffiffi
kr n kL2 6 2 c2 b1 ð2 þ L1 L2 þ L1 L3 Þh: ð3:15Þ
This completes the proof. h

Theorem 3.2. Let s ¼ mh with m a positive integer, then the method (2.10) for index 1 SDDAEs (2.1) is strongly convergent with
order 1/2.

Proof. By (2.10) and (3.1), we can obtain

xnþ1  xðtnþ1 Þ ¼ xn  xðtn Þ þ hh½f ðxnþ1 ; xnmþ1 ; ynþ1 Þ  f ðxðt nþ1 Þ; xðt nþ1  sÞ; yðt nþ1 ÞÞ þ ð1  hÞh½f ðxn ; xnm ; yn Þ
 f ðxðt n Þ; xðtn  sÞ; yðtn ÞÞ þ ½gðxn ; xnm ; yn Þ  gðxðtn Þ; xðt n  sÞ; yðt n ÞÞMW n  r n : ð3:16Þ
Here s ¼ mh and tn  s ¼ tnm . Denote
M n ¼ xn  xðt n Þ;
/fn ¼ f ðxn ; xnm ; yn Þ  f ðxðt n Þ; xðtn  sÞ; yðt n ÞÞ;
ð3:17Þ
/g n ¼ gðxn ; xnm ; yn Þ  gðxðt n Þ; xðt n  sÞ; yðtn ÞÞ;
 n ¼ h/fnþ1 þ ð1  hÞ/fn :
/f
Therefore,
 n þ /g MW n  r n ;
Mnþ1 ¼ M n þ h/f ð3:18Þ
n

which implies

kM nþ1 k2L2 ¼ hM n þ h/f

 n þ /g MW n  rn ; M n þ h/f
n
 n þ /g MW n  r n i
n
h
2 2 2 2 T
 n j h þ jrn j þ 2hM /f n þ 2MT /g MW n  2M T r n
¼ E jMn j þ j/f n n n n
i
 T T  T
þ 2hð/fn Þ /g n MW n  2ð/g n MW n Þ r n  2hð/fn Þ r n þ h/g n MW n ; /g n MW n i: ð3:19Þ

 n ÞT /g MW n jFt  ¼ 0,
Since WðtÞ is a standard s-dimensional Brownian motion, we have E½M Tn /g n MW n jFtn  ¼ 0, E½ð/f n n

E½ð/g n MW n ÞT r n jFtn  ¼ 0, and

     T
2 
kM nþ1 k2L2 ¼ EjM n j2 þ h Ej/f 2 2 2 T T 
n j þ hEj/g n j þ Ejr n j þ 2hE M n /fn  2E M n r n  2hE /fn r n : ð3:20Þ

Similar to (3.10), one has

jyn  yðtn Þj 6 L1 L2 jxn  xðt n Þj þ L1 L3 jxnm  xðt nm Þj: ð3:21Þ
By (3.21) and GLC (2.4), it follows that
j/fn j 6 b1 ½ð1 þ L1 L2 ÞjMn j þ ð1 þ L1 L3 ÞjMnm j: ð3:22Þ
2 2 2
Denote K 1 :¼ b1 ð1 þ L1 L2 Þ and K 2 :¼ b1 ð1 þ L1 L3 Þ. Considering ðhx þ ð1  hÞyÞ 6 hx þ ð1  hÞy , we have
   
 n j2 6 hj/fnþ1 j2 þ ð1  hÞj/fn j2 6 2h K 2 jM nþ1 j2 þ K 2 jM nþ1m j2 þ 2ð1  hÞ K 2 jMn j2 þ K 2 jM nm j2 :
j/f 1 2 1 2

2 1
It is known that xn ; xðtn Þ; xn  xðt n Þ are Ftn -measurable, thus by the fundamental inequality 2xy 6 hx þ h y2 ; 0 < h < 1 and
(3.12), we have
          1 1h i
 T   T   T  2 1 2 2 2 2
E M n r n  ¼ E Etn M n r n  ¼ E Mn Etn ðr n Þ  6 ½hEjMn j þ h EðjðEtn ðr n ÞÞj Þ ¼ hEjMn j þ K 3 h ; ð3:23Þ
2 2
pffiffiffiffiffi
where K 3 ¼ c2 b1 ð2 þ L1 L2 þ L1 L3 Þ. Additionally,
     
2jE M Tn /f
 n j ¼ 2hjE M T /fnþ1 j þ 2ð1  hÞjE M T /fn j
n n

6 K 1 hEjM nþ1 j2 þ ðK 1 ð2  hÞ þ K 2 ÞEjM n j2 þ K 2 hEjM nmþ1 j2 þ K 2 ð1  hÞEjMnm j2 ; ð3:24Þ

j/g n j2 ¼ jgðxn ; xnm ; yn Þ  gðxðtn Þ; xðt n  sÞ; yðt n ÞÞj2

6 b21 ½jM n j þ jMnm j þ jyn  yðt n Þj2
6 2b21 ½ð1 þ L1 L2 Þ2 jM n j2 þ ð1 þ L1 L3 Þ2 jM nm j2  ð3:25Þ
¼ 2K 21 jM n j2 þ 2K 22 jMnm j2 ;
 n ÞT r n Þj 6 hEðj/f
2hjEðð/f  n j2 þ jrn j2 Þ: ð3:26Þ
4014 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021

Substituting (3.23)–(3.26) into (3.20), we can show that

  h  
2 2
kM nþ1 k2L2 6 EjM n j2 þ 2h K 21 EjM n j2 þ K 22 EjM nm j2 þ 4K 23 ð1 þ hÞh þ ðh þ hÞ 2h K 21 EjM nþ1 j2 þ K 22 EjMnþ1m j2
 i
þ 2ð1  hÞ K 21 EjM n j2 þ K 22 EjM nm j2 þ K 1 hhEjM nþ1 j2 þ ðK 1 ð2  hÞ þ K 2 ÞhEjMn j2
2
þ K 2 hhEjM nmþ1 j2 þ K 2 ð1  hÞhEjMnm j2 þ hEjM n j2 þ K 23 h
  h i
2 2
¼ 2K 21 hðh þ hÞ þ K 1 hh EjM nþ1 j2 þ 1 þ 2K 21 ð1  hÞðh þ hÞ þ 2K 21 h þ ðK 1 ð2  hÞ þ K 2 Þh þ h EjM n j2
h i h i
2 2 2
þ 2K 22 hðh þ hÞ þ K 2 hh EjM nmþ1 j2 þ ð5 þ 4hÞK 23 h þ 2K 22 ð1  hÞðh þ hÞ þ 2K 22 h þ K 2 ð1  hÞh EjMnm j2 :
n o
2
1
Set h0 ¼ min 1; 2ð1þ4K ÞK
. When h < h0 , we have 0 < 1  2K 21 hðh þ hÞ  K 1 hh < 1. Noting that h 6 1 and 0 6 h; 1  h 6 1,
1 1 h   i
2
there must exist a sufficiently large constant K such that K 1  2K 21 h h0 þ h0  K 1 hh0 P 1, then
  h  
2 2
1  2K 21 hðh þ hÞ  K 1 hh EjM nþ1 j2 6 1 þ 2K 21 ð1  hÞ þ 2K 22 ðh þ hÞ
  i
2
þ K 1 ð2  hÞ þ 2K 2 þ 1 þ 2K 21 þ 2K 22 h max EjM i j2 þ 9K 23 h ; ð3:27Þ
nm6i6n

1 h  
2
EjM nþ1 j2 6 2
1 þ 2K 21 ð1  hÞ þ 2K 22 ðh þ hÞ
1 2K 21 hðh
þ hÞ  K 1 hh
  i 
2
þ K 1 ð2  hÞ þ 2K 2 þ 1 þ 2K 21 þ 2K 22 h max EM2i þ 9K 23 h
nm6i6n
2     3
2
2 K 21 þ K 22 ðh þ hÞ þ 2K 1 þ 2K 2 þ 1 þ 2K 21 þ 2K 22 h
¼ 41 þ 2
5 max EM2
i
1  2K 21 hðh þ hÞ  K 1 hh nm6i6n

2
9K 23 h
þ 2
1 2K 21 hðh
þ hÞ  K 1 hh
 h   i 
2
6 1 þ K 6 K 1 þ K 22 þ 2K 1 þ 2K 2 þ 1 h max EjM i j2 þ 9KK 23 h :
2
ð3:28Þ
nm6i6n
 
Let a ¼ K 6K 21 þ 6K 22 þ 2K 1 þ 2K 2 þ 1 ; b ¼ 9KK 23 . Clearly, a; b are constants independent of h. Then (3.28) turns to be
2
EjM nþ1 j2 6 ð1 þ ahÞ max EjM i j2 þ bh ;
nm6i6n

which implies

EjM nþ1 j2 6 ð1 þ ahÞn max EjM i j2 þ ðeaT  1Þba1 h:

m6i60
1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Therefore, we have kxnþ1  xðtnþ1 ÞkL2 6 C 1 h2 , where C 1 ¼ ðeaT  1Þba1 . Furthermore, by (3.12), there exists a non-negative
constant C 2 such that
1
kynþ1  yðt nþ1 ÞkL2 6 L1 L2 kM nþ1 kL2 þ L1 L3 kM nþ1m kL2 6 C 2 h2 :
From here, we can conclude that there exists a non-negative constant C such that
1
max ðkxn  xðtn ÞkL2 þ kyn  yðt n ÞkL2 Þ 6 Ch2 ; ð3:29Þ
06n6N

where C 1 ; C 2 ; C are independent of h, which proves the theorem. h

Remark 3.1. Applying Theorems 3.1 and 3.1 to the case of systems without delay, we can also obtain that the stochastic
theta method is strongly consistent and convergent with order 1/2, which corresponds to the results in [20]. Therefore,
our conclusions of convergence can be seen as an extension of the corresponding result in [20].

Theorem 3.3. Let s ¼ ðm  dÞh with m a positive integer and d 2 ½0; 1Þ, then stochastic theta method (2.10) with linear interpo-
lation (2.11) is strongly convergent with order 12.

Proof. By (3.1) and (2.12), it is easy to see that

xnþ1  xðtnþ1 Þ ¼ xn  xðtn Þ þ hh½f ðxnþ1 ; xnþ1 ; ynþ1 Þ  f ðxðtnþ1 Þ; xðtnþ1  sÞ; yðtnþ1 ÞÞ þ ð1  hÞh½f ðxn ; xn ; yn Þ
 f ðxðtn Þ; xðt n  sÞ; yðt n ÞÞ þ ½gðxn ; xn ; yn Þ  gðxðt n Þ; xðtn  sÞ; yðt n ÞÞMW n  rn :
Denote
 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021 4015

M n ¼ xn  xðt n Þ;
~ n ¼ f ðxn ; xn ; y Þ  f ðxðtn Þ; xðt n  sÞ; yðt n ÞÞ;
/f n

~ ¼ gðxn ; xn ; y Þ  gðxðt n Þ; xðtn  sÞ; yðt n ÞÞ;

/g n n

/f ~ nþ1 þ ð1  hÞ/f
^ n ¼ h/f ~ n;

then

^ n þ /g
Mnþ1 ¼ M n þ h/f ~ MW n  r n : ð3:30Þ
n

Similar to (3.20), we have

   
2 ^
kM nþ1 k2L2 ¼ EjM n j2 þ h Ej/f 2 ~ 2 2 T^ T ^ T
n j þ hEj/g n j þ Ejr n j þ 2hE M n /fn  2E M n r n  2hEðð/fn Þ r n Þ: ð3:31Þ

By the Hölder and Cauchy inequalities, we can show that

  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
^ n j 6 2h EjMn j2 Ej/f
2hjE M Tn /f ^ n j2 6 h EjMn j2 þ Ej/f
^ n j2 ;
qffiffiffiffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
^ n ÞT r n Þj 6 2h Ejr n j2 Ej/f
2hjEðð/f ^ n j2 6 h Ejrn j2 þ Ej/f
^ n j2 :

Therefore, a combination of (3.23) and (3.31) implies

2 ~ j2 þ ð1 þ hÞEjrn j2 þ K 2 h2 :
EjM nþ1 j2 6 ð1 þ 2hÞEjM n j2 þ ðh þ 2hÞEj/f
^ n j2 þ hEj/g
n 3 ð3:32Þ

Using (2.11), we have

jxn  xðt n  sÞj ¼ jdxnþ1m þ ð1  dÞxnm  xðtn  sÞj

6 djM nmþ1 j þ ð1  dÞjM nm j þ djxðt nþ1m Þ  xðt n  sÞj þ ð1  dÞjxðt nm Þ  xðtn  sÞj; ð3:33Þ
then

~ n j 6 b ½ð1 þ L1 L2 ÞjMn j þ ð1 þ L1 L3 Þ½djM nmþ1 j þ ð1  dÞjM nm j þ djxðt nþ1m Þ  xðtn  sÞj þ ð1  dÞjxðt nm Þ  xðt n  sÞj
j/f 1

¼ K 1 jMn j þ K 2 ½djMnmþ1 j þ ð1  dÞjM nm j þ djxðt nþ1m Þ  xðtn  sÞj þ ð1  dÞjxðtnm Þ  xðt n  sÞj:

By Lemma 3.1 and the above inequality, we have

~ n j2 6 E½K 1 jMn j þ K 2 ½djM nmþ1 j þ ð1  dÞjM nm j þ djxðt nþ1m Þ  xðtn  sÞj þ ð1  dÞjxðt nm Þ  xðt n  sÞj2
Ej/f
6 2K 21 EjMn j2 þ 8K 22 d2 EjM nþ1m j2 þ 8K 22 ð1  dÞ2 EjMnm j2 þ 8K 22 d2 Ejxðt nþ1m Þ  xðtn  sÞj2
þ 8K 22 ð1  dÞ2 Ejxðt nm Þ  xðtn  sÞj2
6 2K 21 EjMn j2 þ 8K 22 d2 EjM nþ1m j2 þ 8K 22 ð1  dÞ2 EjMnm j2 þ 8K 22 c2 ½d2 ð1  dÞ þ dð1  dÞ2 h;
hence

^ n j2 6 hEj/f
Ej/f ~ nþ1 j2 þ ð1  hÞEj/f ~ n j2
h i
6 2K 21 EjM nþ1 j2 þ 8K 22 d2 EjMnþ2m j2 þ 8K 22 ð1  dÞ2 EjMnþ1m j2 h
h i
þ 2K 21 EjM n j2 þ 8K 22 d2 EjM nþ1m j2 þ 8K 22 ð1  dÞ2 EjM nm j2 ð1  hÞ þ 8K 22 c2 dð1  dÞh: ð3:34Þ

Noting that (3.21), we have

~ j2 ¼ Ejgðxn ; xn ; y Þ  gðxðtn Þ; xðt n  sÞ; yðt n ÞÞj2 6 2b2 ð1 þ L1 L2 Þ2 EjM n j2 þ 2b2 ð1 þ L1 L3 Þ2 Ejxn  xðt n  sÞj2
Ej/g n n 1 1

6 2K 21 EjM n j2 þ 8K 22 d2 EjM nþ1m j2 þ 8K 22 ð1  dÞ2 EjM nm j2 þ 8K 22 c2 dð1  dÞh: ð3:35Þ

It therefore follows that
2 2 2
EjM nþ1 j2 6 2K 21 hðh þ 2hÞEjM nþ1 j2 þ 8K 22 hd2 ðh þ 2hÞEjMnþ2m j2 þ ½1 þ 2h þ 2K 21 h þ 2K 21 ð1  hÞðh þ 2hÞEM2n
h i
2 2
þ ð5 þ 4hÞK 23 h þ 8K 22 ½ð1  dÞ2 h þ d2 ð1  hÞðh þ 2hÞ þ 8K 22 d2 h EjM nþ1m j2 þ 8K 22 ð1  dÞ2 h½1 þ ð1
2
 hÞð2 þ hÞEjM nm j2 þ 8K 22 c2 dð1  dÞð3 þ hÞh : ð3:36Þ
Next we consider the two cases m ¼ 1 and m P 2 separately.
4016 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021

1. In the case of m ¼ 1, (3.36) turns to be

 
2 2 2
EjM nþ1 j2 6 2K 21 þ 8K 22 d2 hðh þ 2hÞEjMnþ1 j2 þ ð5 þ 4hÞK 23 h þ 8K 22 c2 dð1  dÞð3 þ hÞh
h   h i i
2
þ 1 þ 2 þ 2K 21 þ 8K 22 ðd2 þ ð1  dÞ2 Þ h þ 8K 22 ðd2 ð1  hÞ þ ð1  dÞ2 Þ þ 2K 21 ð1  hÞ ðh þ 2hÞ max EjMi j2 ;
nm6i6n

then
    h     i
2 2
1  2K 21 þ 8K 22 d2 hðh þ 2hÞ EjMnþ1 j2 6 1 þ 2 þ 2K 21 þ 8K 22 h þ 8K 22 þ 2K 21 ð1  hÞ ðh þ 2hÞ max EjMi j2
nm6i6n
2 2
þ ð5 þ 4hÞK 23 h þ 8K 22 c2 dð1  dÞð3 þ hÞh :
ð3:37Þ
2. In the case of m P 2, we can obtain from (3.36)
  h     i
2 2
1  2K 21 hðh þ 2hÞ EjM nþ1 j2 6 1 þ 2 þ 2K 21 þ 8K 22 h þ 2K 21 ð1  hÞ þ 8K 22 ðh þ 2hÞ max EjM i j2
nm6i6n
2 2
þ ð5 þ 4hÞK 23 h þ 8K 22 c2 dð1  dÞð3 þ hÞh : ð3:38Þ
Therefore, either (3.37) or (3.38) is just different from the inequality (3.27) in the coefficients. Using the same way as in the
proof of Theorem 3.2, we can show the assertion (3.29). h

Remark 3.2. Our work of this section is motivated by a recent meeting report [21], which discusses the convergence of
Euler–Maruyama method for a class of stochastic differential algebraic system of index 1 with delay.

Remark 3.3. Applying Theorems 3.1–3.3 to stochastic delay differential equations, we also obtain that the stochastic theta
method is convergent for SDDEs. Therefore our results generalize the conclusion of Theorem 7 in [3].

4. Stability analysis

In this section, we aim at the stability properties of analytical and numerical solutions for (2.1).
Various stability properties of numerical methods for SDDEs have been studied (see [4] for example), while not much
work for that of SDAEs can be found (see [20]). For the purpose of stability analysis, we further assume that uð0; 0; 0Þ ¼ 0.
Let us state a theorem which provides us a criterion on the asymptotically stability in the mean-square sense (abbreviated
as MS-stability) of the exact solution for Eq. (2.1).
Theorem 4.1. If f,g satisfy
8
>
> xT f ðx; 0; 0Þ 6 a jxj2 ;
>
>
>
< jf ðx; m; y1 Þ  f ðx1 ; m1 ; yÞj 6 a0 jx  x1 j þ a1 jm  m1 j þ a2 jy  y1 j;
ð4:1Þ
>
> jgðx; m; yÞj2 6 b 0 jxj2 þ b 1 jmj2 þ b
2 jyj2 ;
>
>  
>
: 2a > 2a1 þ b 0 þ b 1 þ 2a2 L1 ðL2 þ L3 Þ þ 2b 2 L2 L2 þ L2 ;
1 2 3

for 8x; x1 ; v ; v 1 2 Rd ; y; y1 2 Rl , where a 0 ; b

; a0 ; a1 ; a2 ; b 1 ; b
2 are non-negative constants, then the solution of index 1 SDDAEs (2.1) is
mean-square stable.
We will prove this conclusion by considering the equivalent form (2.8) of (2.1). To this end, we first state a stability result
concerning SDDEs.
Lemma 4.1. [12]. For SDDEs (2.9), if f,g satisfy
(
2xT ~f ðt; x; mÞ þ jg~ðt; x; mÞj2 6 k0 jxj2 þ k1 jmj2
k0 > k1
where t P 0; x; m 2 Rd ; k0 ; k1 > 0 are constants, then the zero solution of SDDEs (2.9) is mean-square stable.
Proof of Theorem 4.1.
For the equivalent form (2.8) of (2.1), considering inequality jUðx; mÞj 6 L1 L2 jxj þ L1 L3 jmj, we have

2xT f ðt; x; m; Uðx; mÞÞ þ jgðt; x; m; Uðx; mÞÞj2 6 2xT jf ðt; x; m; Uðx; mÞÞ  f ðt; x; 0; 0Þ þ f ðt; x; 0; 0Þj þ jgðt; x; m; Uðx; mÞÞj2
jxj2 þ b
6 2xT ða1 m þ a2 Uðx; mÞÞ  2a 0 jxj2 þ b
1 jmj2 þ b
2 jUðx; mÞj2

6 ða1 þ 2a2 L1 L2 þ a2 L1 L3 Þjxj2 þ ða1 þ a2 L1 L3 Þjmj2

 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021 4017

jxj2 þ b
 2a 0 jxj2 þ b
1 jmj2 þ b2 jUðx; mÞj2

6 ða1 þ 2a2 L1 L2 þ a2 L1 L3 þ b0  2a  þ 2b2 L2 L2 Þjxj2

1 2
2 L2 L2 Þjmj2 ¼ k0 jxj2 þ k1 jmj2 ;
1 þ a2 L1 L3 þ 2b
þ ða1 þ b 1 3

where k0 ¼ 2a   a1  b 0  2a2 L1 L2  a2 L1 L3  2b 2 L2 L2 > 0, k1 ¼ a1 þ b 2 L2 L2 . Using (4.1), we can easily verify

1 þ a2 L1 L3 þ 2b
1 2 1 3
k0 > k1 . Because jyðtÞj 6 L1 L2 jxðtÞj þ L1 L3 jxðt  sÞj, so ðxðtÞ; yðtÞÞ is mean-square stable. This completes the proof of Theorem
4.1.
Remark 4.1. Apply Theorem 4.1 to SDDEs, which implies a2 ¼ b 2 ¼ 0. Then the last condition in (4.1) turns to be
0 þ b
 > 2a1 þ b
2a 1 . This corresponds to results in [12,13].

Next we discuss MS-stability of the stochastic theta method for (2.1). For brevity, we assume the Brownian motion is 1-D.
Defintion 4.1. The numerical method (2.10) is said to be MS-stable if there exists a positive constant h0 such that for all
h 2 ð0; h0 Þ, we have

lim Ejxn j2 ¼ 0:
n!1

Theorem 4.2. Assume that all the conditions of Theorem 4.1 hold, and s ¼ mh with m a positive integer, then the stochastic theta
method (2.10) for index 1 SDDAEs is MS-stable.

Proof. Noting that s ¼ mh, we have

xnþ1  hf ðxnþ1 ; xnþ1m ; ynþ1 Þh ¼ xn þ ð1  hÞf ðxn ; xnm ; yn Þh þ gðxn ; xnm ; yn ÞMW n ;
then
2 2
jxnþ1 j2 ¼ jxn j2 þ 2hhxTnþ1 f ðxnþ1 ; xnmþ1 ; ynþ1 Þ  h h2 jf ðxnþ1 ; xnmþ1 ; ynþ1 Þj2 þ h ð1  hÞ2 jf ðxn ; xnm ; yn Þj2
2
þ jgðxn ; xnm ; yn ÞMW n j þ 2hð1  hÞxTn f ðxn ; xnm ; yn Þ þ 2xTn gðxn ; xnm ; yn ÞMW n
T
þ 2hð1  hÞf ðxn ; xnm ; yn Þ gðxn ; xnm ; yn ÞMW n : ð4:2Þ

Using (3.8) and (3.9), we can obtain

jyn j 6 L1 L2 jxn j þ L1 L3 jxnm j; ð4:3Þ

therefore
T T
hjxn j2
2xTn f ðxn ; xnm ; yn Þh 6 2hxn ½f ðxn ; xnm ; yn Þ  f ðxn ; 0; 0Þ þ f ðxn ; 0; 0Þ 6 2hxn ½a1 jxnm j þ a2 jyn j  2a
T
hjxn j2
6 2hxn ½ða1 þ a2 L1 L3 Þjxnm j þ a2 L1 L2 jxn j  2a
 þ 2a2 L1 L2 Þhjxn j2 þ ða1 þ a2 L1 L3 Þhjxnm j2 :
6 ða1 þ a2 L1 L3  2a ð4:4Þ
By (4.1), we have
   
jgðxn ; xnm ; yn Þj2 6 b
0 jxn j2 þ b
1 jxnm j2 þ b
2 jy j2 6 b
n
2 L2 L2 jxn j2 þ b
 0 þ 2b
1 2
2 L2 L2 jxnm j2 ;
 1 þ 2b
1 3
ð4:5Þ
e 2 jxn j2 þ 2 K
jf ðxn ; xnm ; yn Þj2 6 ða0 jxn j2 þ a1 jxnm j2 þ a2 jyn j2 Þ2 6 2 K e 2 jxnm j2 ;
1 2

e 1 ¼ a0 þ a2 L1 L2 , K
where K e 2 ¼ a1 þ a2 L1 L3 . Because EðMW n jFt Þ ¼ 0, E½ðMW n Þ2 jFt  ¼ h, and xn ; xnm are Ft -measurable, we
n n n

can compute that


E xTn gðxn ; xnm ; yn ÞMW n jFtn ¼ 0;
h i
E f ðxn ; xnm ; yn ÞT gðxn ; xnm ; yn ÞMW n jFtn ¼ 0:

Substituting (4.4), (4.5) and the above inequalities into (4.2), then taking expectation on both sides, we have
2
Ejxnþ1 j2 6 Ejxnþ1 j2 þ h h2 Ejf ðxnþ1 ; xnmþ1 ; ynþ1 Þj2
h i
6 Ejxn j2 þ hh ða1 þ a2 L1 L3 þ 2a2 L1 L2  2a ÞEjxnþ1 j2 þ ða1 þ a2 L1 L3 ÞEjxnþ1m j2
h i     
2
þ 2h ð1  hÞ2 K e 2 Ejxn j2 þ K
e 2 Ejxnm j2 þ h b  0 þ 2b2 L2 L2 Ejxn j2 þ b1 þ 2  arb L2 L2 Ejxnm j2
1 2 1 2 2 1 3
h i
þ hð1  hÞ ða1 þ a2 L1 L3 þ 2a2 L1 L2  2a ÞEjxn j2 þ ða1 þ a2 L1 L3 ÞEjxnm j2 ;

which implies
4018 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021

h  
ÞÞEjxnþ1 j2 6 1 þ hð1  hÞða1 þ a2 L1 L3 þ 2a2 L1 L2  2a
ð1  hhða1 þ a2 L1 L3 þ 2a2 L1 L2  2a Þ þ h b 2 L2 L2
 0 þ 2b
1 2
i
2
þ 2h ð1  hÞ2 Ke 2 Ejxn j2 þ ða1 þ a2 L1 L3 ÞhhEjxnþ1m j2
1
h   i
þ 2Ke 2 h2 ð1  hÞ2 þ b 2 L2 L2 h þ ða1 þ a2 L1 L3 Þð1  hÞh Ejxnm j2 :
 1 þ 2b
2 1 3

ð4:6Þ
Denote
 
P ¼ 1 þ ð1  hÞða1 þ a2 L1 L3 þ 2a2 L1 L2  2a
Þh þ b 0 þ 2b e 2 h2 ;
2 L2 L2 h þ 2ð1  hÞ2 K
1 2 1

Q ¼ ða1 þ a2 L1 L3 Þhh;
 
e 2 h2 ð1  hÞ2 þ b
R ¼ 2K 2 L2 L2 h þ ða1 þ a2 L1 L3 Þð1  hÞh;
 1 þ 2b
2 1 3

then it obeys that

ÞhÞEjxnþ1 j2 6 PEjxn j2 þ QEjxnþ1m j2 þ REjxnm j2 :
ð1  hða1 þ a2 L1 L3 þ 2a2 L1 L2  2a ð4:7Þ
Since (4.1) holds, we have a1 þ a2 L1 L3 þ 2a2 L1 L2  2a  < 0 and 1  hhða1 þ a2 L1 L3 þ 2a2 L1 L2  2a
Þ > 0. Denote A ¼ 2ð1
2 e2   2 2 2 2
hÞ K 1 , p ¼ b0 þ 2b2 L1 L2 þ ð1  hÞða1 þ a2 L1 L3 þ 2a2 L1 L2  2aÞ. IfB  4A < 0, then P ¼ 1 þ Bh þ Ah > 0. Otherwise, when
Bffiffiffiffiffiffiffiffiffiffi 
2
h < jBj 2AB 4A ; P > 0 holds. Clearly, Q ; R > 0. Hence (4.7) can be turned to
PþQ þR
Ejxnþ1 j2 6 max Ejxi j2 ;
1  S nm6i6n
Þ < 0. Once P þ Q þ R < 1  S, we have lim Ejxn j2 ¼ 0. Consequently, we discuss the
where S ¼ hhða1 þ a2 L1 L3 þ 2a2 L1 L2  2a
n!1
inequality P þ Q þ R < 1  S, which is equivalent to
     
2ð1  hÞ2 K e 2 h þ 2a1 þ b
e2 þ K 0 þ b 2 L2 L2 þ L2  2a
1 þ 2a2 L1 ðL2 þ L3 Þ þ 2b  < 0:
1 2 1 2 3

0 b
2a1 b
2a 2 L2 ðL2 þL2 Þ
1 2a2 L1 ðL2 þL3 Þ2b
When h < 1 2 3
, the above inequality is valid.
2ð1hÞ2 ðe
K 21 þe
K 22 Þ

To sum up, let all the conditions of Theorem 4.1 hold, and choose

8
>
> C ; B > 0 or B2  4A < 0;
< 2ð1hÞ2 eK 21 þeK 22
h0 ¼  pffiffiffiffiffiffiffiffiffiffi
>
> C jBj B2 4A
: min ; ; otherwise;
2ð1hÞ2 ðe K 21 þe
2A
K 22 Þ
 
0  b
  2a1  b
where C ¼ 2a 2 L2 L2 þ L2 , then for all h < h0 , the stochastic theta method for index 1
1  2a2 L1 ðL2 þ L3 Þ  2b
1 2 3
SDDAEs is MS-stable. h

Remark 4.2. If we remove the restriction s ¼ mh, then we can similarly prove that the stochastic theta method with linear
interpolation (2.11) is MS-stable.

5. Numerical examples

We use the equation

8
< dxðtÞ ¼ ½axðtÞ þ bxðt  1Þ þ yðtÞdt þ ½cxðtÞ þ dxðt  1ÞdWðtÞ;
> t P 0;
0 ¼ l sin xðt  1Þ þ qxðtÞ  yðtÞ; t P 0; ð5:1Þ
>
:
xðtÞ ¼ t þ 1; yðtÞ ¼ l sin t þ qðt þ 1Þ; t 2 ½1; 0;
as a test equation, where a; b; c; d; l; q are constants. This equation can be called the semi-explicit index-1 SDDAEs. The ex-
plicit solution on t 2 ½0; 1 is given by
 Z t Z t 
xðtÞ ¼ /t;0 1 þ /1
s;0 ðbs þ l sin s  cdsÞds þ ds/1
s;0 dWðsÞ ;
0 0
 

c2
/t;0 ¼ exp a þ q  t þ cWðtÞ :
2
We obtain the explicit solution on the second interval [1,2] by using the above solution as the new initial function. In fact, on
every interval the computing method is the same as the method for stochastic differential equation. We compute the integral
in the solution by the linear interpolation, and we choose the stepsize as 1/1024.
 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021 4019

First, let us consider the influence of parameters for (5.1). In all figures t n denotes the mesh-point. We have used the set of
coefficients I: a ¼ 1; b ¼ 0; c ¼ 1; d ¼ 0; l ¼ 0; q ¼ 0, II: a ¼ 1; b ¼ 1; c ¼ 1; d ¼ 1; l ¼ 0; q ¼ 0, III: a ¼ 1; b ¼ 1; c ¼ 1; d ¼ 1;
l ¼ 0:1; q ¼ 10. We all use h ¼ 0:2; h ¼ 25 from Figs. 1–3. The data used in these figures are obtained by the mean of data by
1000 trajectories. One may consider (5.1) with coefficients I to be a linear stochastic ordinary differential equation. When
2
a <  c2 , the solution of this equation is mean-square stable [12]. Case II for (5.1) can be seen as a linear stochastic delay dif-
2
ferential equation. When a < jbj  ðjcjþjdjÞ
2
, the solution of this equation is mean-square stable [8]. While case III for (5.1) can
be considered as SDDAEs. From Figs. 1–3, we observe a change from stable to unstable then stable, varying from the
coefficients.

0.8

0.6

Xn 0.4

0.2

−0.2
0 1 2 3 4 5 6
tn

Fig. 1. Simulations with coefficients I for (5.1).

10

4
Xn
2

−2

−4
0 0.5 1 1.5 2 2.5
tn

Fig. 2. Simulations with coefficients II for (5.1).

1
0.9
0.8
0.7
0.6
Xn 0.5
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6
tn

Fig. 3. Simulations with coefficients III for (5.1).

4020 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021

1 exact solution
0.9 Euler−Maruyama

0.8
0.7
0.6
Xn 0.5
0.4
0.3
0.2
0.1
0
0 0.5 1 1.5 2 2.5 3 3.5 4
tn

Fig. 4. The stable analytical solution and numerical solution with coefficients IV for (5.1).

2.5
exact solution
bakward−Euler

Xn 1.5

0.5
0 0.5 1 1.5 2 2.5 3 3.5 4
tn

Fig. 5. The unstable analytical solution and numerical solution with coefficients V for (5.1).

Next, we consider the theoretical order of strong convergence. We have used the coefficients IV: a ¼ 8; b ¼ 2;
c ¼ 0; d ¼ 1; l ¼ 1; q ¼ 1, and V: a ¼ 4; b ¼ 1; c ¼ 0; d ¼ 3; l ¼ 1; q ¼ 2. We choose the stepsize h ¼ 25 and h ¼ 0 for Fig. 4,
h ¼ 1 for Fig. 5 separately. The mean-square error EjxðTÞ  X N j2 at the final time T ¼ 4 is estimated in the following way.
A set of 20 blocks each containing 100 outcomes ðW ij : 1 6 i 6 20; 1 6 j 6 100Þ are simulated and for each block the
estimator

1 X100
ei ¼ jxðT; W ij Þ  X N ðW ij Þj2
100 j¼1

is formed. In Table 1, e denotes the mean of this estimator, which is itself estimated in the usual way:
1 X
20
ei :
20 i¼1

The numerical results and the explicit solution are compared in Figs. 4 and 5, the results show that numerical solutions con-
verge to explicit solution.
Finally, we check whether the stability conditions (4.1) is valid. We use the coefficients VI: a ¼ 4; b ¼
1; c ¼ 0; d ¼ 1; l ¼ 0:2; q ¼ 1. For the SDDAEs, we can choose L1 ¼ 1; L2 ¼ q ¼ 1; L3 ¼ l ¼ 0:2; a 0 ¼ b
 ¼ a ¼ 4; b 2 ¼ 0; b
1 ¼ 1.
By validating conditions (4.1), we ensure that the solution of this equation is mean-square stable. The data is obtained by
1
P1000 2 5
the mean-square of 1000 trajectories, that is, W i : 1 6 i 6 1000; X n ¼ 1000 i¼1 jX n ðW i Þj . Choosing h ¼ 0:5; h ¼ 2 , the sim-
ulation results are shown in Fig. 6. Clearly, the STM reveals the stability of the solution.

Table 1
The convergence of BE method applied to (5.1) with IV.

Stepsize 1 1 1 1 1
8 16 32 64 128

e 2.9806-04 1.5971e04 7.2109e05 4.045e05 2.3559e05

 X. Qu et al. / Applied Mathematics and Computation 215 (2010) 4008–4021 4021

1
0.9
0.8
0.7
0.6
Xn 0.5
0.4
0.3
0.2
0.1
0
0 5 10 15
tn

Fig. 6. Simulations with coefficients VI for (5.1).

Acknowledgement

The authors wish to thank the anonymous referees for their kind and valued comments. This work was supported by NSF
of China (No. 10971077) and by Program for NCET, the State Education Ministry of China.

References

[1] M. Arnold, Stability of numerical methods for differential-algebraic equations of high index, Appl. Numer. Math. 13 (1993) 5–14.
[2] A. Bellen, M. Zennaro, Numerical Methods for Delay Differential Equations, Oxford University Press, UK, 2003.
[3] E. Buckwar, Introduction to the numerical analysis of stochastic delay differential equations, J. Comput. Appl. Math. 125 (2000) 297–307.
[4] E. Buckwar, One-step approximations for stochastic functional differential equations, Appl. Numer. Math. 56 (2006) 667–681.
[5] H. Brunner, Collocation Methods for Volterra Integral and Related Functional Differential Equations, Cambridge University Press, UK, 2004.
[6] S.L. Campbell, V.H. Linh, Stability criteria for differential-algebraic equations with multiple delays and their numerical solutions, Appl. Math. Comput.
208 (2009) 397–415.
[7] S.L. Campbell, Nonregular 2D descriptor delay systems, IMA J. Math. Cont. Inf. 12 (1995) 57–67.
[8] W. Cao, M. Liu, Z. Fan, MS-stability of the Euler–Maruyama method for stochastic differential delay equations, Appl. Math. Comput. 159 (2004) 127–
135.
[9] D.J. Higham, An algorithmic introduction to numerical simulation of stochastic differential equations, SIAM Rev. 43 (2001) 525–546.
[10] A. Jenten, P.E. Kloeden, A. Neuenkirch, Pathwise approximation of stochastic differential equations on domains: higher order convergence rates
without global Lipschitz coefficients, Numer. Math. 112 (2009) 41–64.
[11] P.E. Kloden, E. Platen, Numerical Solution of Stochastic Differential Equations, Springer, Berlin, 1992.
[12] X. Mao, Stochastic Differential Equations and Applications, Horwood, England, 1997.
[13] X. Mao, S. Sabanis, Numerical solutions of stochastic differential delay equations under local Lipschitiz condition, J. Comput. Appl. Math. 151 (2003)
215–227.
[14] X. Mao, Exponential Stability of Stochastic Differential Equations, Marcel Dekker, New York, 1994.
[15] E. Platen, An introduction to numerical methods for stochastic differential equations, Acta Numer. 8 (1999) 197–246.
[16] G. Reißig, W.S. Martinson, P. Barton, Differential-algebraic equations of index 1 may have an arbitrarily high structural index, SIAM J. Sci. Comput. 21
(2000) 1987–1990.
[17] Y. Ren, N. Xia, A note on the existence and uniqueness of the solution to neutral stochastic functional differential equations with infinite delay, Appl.
Math. Comput. 214 (2009) 457–461.
[18] O. Schein, G. Denk, Numerical solution of stochastic differential-algebraic equations with applications to transient noise simulation of microelectronic
circuits, J. Comput. Appl. Math. 100 (1998) 77–92.
[19] L. Torelli, A sufficient condition for GRN-stability for delay differential equations, Numer. Math. 59 (1991) 311–320.
[20] R. Winkler, Stochastic differential algebraic equations of index 1 and applications in circuit simulation, J. Comput. Appl. Math. 157 (2003) 477–505.
[21] F. Xiao, C. Zhang, Euler–Maruyama method for a class of stochastic differential algebraic system with delay, in: The 11th Conference on Numerical
Methods of Differential Equations, Qingdao, China, 2008.
[22] W. Zhu, L. Petzold, Asymptotic stability of linear delay differential-algebraic equations and numerical methods, Appl. Numer. Math. 24 (1997) 247–
264.