Home Convergence rate of the truncated Euler-Maruyama method for highly nonlinear neutral stochastic differential equations with time-dependent delay
Article Open Access

Convergence rate of the truncated Euler-Maruyama method for highly nonlinear neutral stochastic differential equations with time-dependent delay

  • Aleksandra M. Petrović EMAIL logo
Published/Copyright: August 14, 2024

Abstract

This article can be considered as a continuation of Petrović and Milošević [The truncated Euler-Maruyama method for highly nonlinear neutral stochastic differential equations with time-dependent delay, Filomat 35 (2021), no. 7, 2457–2484], where the authors established the L q -convergence of the truncated Euler-Maruyama (EM) method for neutral stochastic differential equations with time-dependent delay under the Khasminskii-type condition. However, the convergence rate of the method has not been studied there, which is the main goal of this article. Also, there are some restrictions on the truncated coefficients of the considered equations, and these restrictions sometimes might force the step size to be so small that the application of the truncated EM method would be limited. Therefore, the convergence rate without these restrictions will be considered in this article. Moreover, one of the sufficient conditions for obtaining the main result of this article, which is related to Lipschitz constants for the neutral term and delay function, is weakened. In that way, some of the results of the cited article are generalized. The main result of this article is proved by employing two conditions related to the increments to the coefficients and the neutral term of the equations under consideration, among other conditions. The main theoretical result is illustrated by an example.

MSC 2010: 60H10; 60H35

1 Introduction and preliminary results

It is well known that many complex dynamical systems depend not only on their present but also on their past states. So, neutral stochastic differential equations with time-dependent delay have become a powerful tool for modeling these systems [15]. The structure of these equations is specific, so it is very difficult to obtain their explicit solutions. Therefore, various numerical methods have taken on important role for generating corresponding approximate solutions of the observed equations such as those from the articles [616], among other publications.

Due to its convenience for computations and implementation, the Euler-Maruyama (EM) method has stood out from other methods, and as such it has often been used to obtain approximate solutions for different types of stochastic differential equations [1721]. However, article by Hutzenthaler et al. [22] reveals a large class of ordinary stochastic differential equations with superlinearly growing coefficients for which the EM method cannot be convergent in the strong and numerically weak L p -sense at a finite time, where p 1 . In particular, due to the advantages of explicit methods, Mao [10] developed a new explicit method called the truncated EM method. In another follow-up article [7], they investigated the convergence rates of the method under some additional conditions. It was a motivation for many authors to further investigate this topic. For example, in [23], the strong convergence of the truncated EM method was established for neutral stochastic differential equations with constant delay and Markovian switching. Also, the main result of this article [11] is establishing a strong convergence of the truncated EM scheme for neutral stochastic differential equations with constant delay, driven by Brownian motion and pure jumps, respectively.

On the other hand, the L q -convergence for q [ 2 , p ) and p 2 of the truncated EM method is considered in [24], and the convergence result was in the asymptotic form without determining the convergence rate. So, one of the main goals in this article is to find the sufficient conditions for determining the convergence rate of the truncated EM method for highly nonlinear neutral stochastic differential equations with time-dependent delay in a finite time interval. More precisely, we will discuss the rate of L q -convergence of the truncated EM method for q > 2 , and we will show that the order of L q -convergence can be arbitrarily close to q 2 . It should be pointed out that certain conditions in [24] might force the step size to be so small that the application of the truncated EM method would be limited. Moreover, in that article, the Lipschitz constant of the neutral term is from the interval ( 0 , 1 3 ) , which restricts the applicability of the obtained results. We will establish the convergence rate without these restrictions, i.e., we will weaken the assumptions under which the assertions of Lemmas 2.1 and 2.2 from [24] hold, which will be given below for convenience of the reader. In order to overcome the difficulties due to the removal of these restrictions, as well as those caused by the presence of the neutral term and delay function, some new mathematical techniques have been developed, which differ significantly from those used in [24], [11], and [23]. It should be emphasized that in [11] and [23], the authors introduced the Khasminski-type condition for the truncated coefficients of the equations under the consideration, while in the article [24], it is proved that, under certain assumptions, this condition holds. By weakening restrictions of the step size Δ , we increase the scope of Δ such that we can simulate the corresponding numerical solutions with a smaller number of steps. These simulations are faster compared to those with a greater number of steps. Moreover, by weakening the restrictions on the Lipschitz constants for the neutral term and delay function, we extend a class of neutral stochastic differential equations with time-dependent delay for which the results of the article [24] hold, and for that extended class of the equations, we will obtain the main results of this article. It should be emphasized that the numerical method under consideration is explicit, so it does not require additional conditions of the existence and uniqueness of the corresponding numerical solution that also influences positively the simplicity and speed of simulating algorithms.

Besides the previously mentioned novelty aspects of the article, it should be stressed that two more conditions, related to the increments of the coefficients and neutral term of the equations under consideration, are imposed. These conditions are essential for proving the L q convergence of the truncated EM method, and their applications represent the novelty of comparing the cited results. At the end, we give an example to illustrate the theoretical results.

Since the results in this article represent a continuation of the work from [24], for convenience, we will use throughout the whole article the same notation as in the cited article. In that sense, it is assumed that all random variables and processes are defined on a filtered probability space ( Ω , , { t } t 0 , P ) with a filtration { t } t 0 satisfying the usual conditions. Let B ( t ) = ( B 1 ( t ) , B 2 ( t ) , , B m ( t ) ) T , t 0 be an m -dimensional standard Brownian motion, t -adapted, and independent of 0 . Let the Euclidean norm be denoted by and, for simplicity, trace [ A T A ] = A 2 for matrix A , where A T is the transpose of a vector or a matrix. Moreover, for two real numbers a and b , we use a b = max ( a , b ) and a b = min ( a , b ) .

For a given τ > 0 , let C ( [ τ , 0 ] ; R d ) be the family of continuous functions φ from [ τ , 0 ] to R d , equipped with the supremum norm φ = sup τ θ 0 φ ( θ ) . Also, denote by C 0 b ( [ τ , 0 ] ; R d ) the family of bounded, 0 -measurable, C ( [ τ , 0 ] ; R d ) -valued random variables.

Let δ : [ 0 , + ) [ 0 , τ ] be the delay function, which is Borel-measurable. We introduce the following autonomous neutral stochastic differential equation with time-dependent delay:

(1) d [ x ( t ) u ( x ( t δ ( t ) ) ) ] = f ( x ( t ) , x ( t δ ( t ) ) ) d t + g ( x ( t ) , x ( t δ ( t ) ) ) d B ( t ) , t 0 ,

satisfying the initial condition

(2) x 0 = ξ = { ξ ( θ ) , θ [ τ , 0 ] } ,

where the functions

f : R d × R d × R + R d , g : R d × R d R d × m , and u : R d R d ,

are all Borel measurable and x ( t ) is a d -dimensional state process. The initial condition ξ is supposed to be a C 0 b ( [ τ , 0 ] ; R d ) -valued random variable.

In order to prove main results of this article, we will use the hypotheses from [24] that are listed below.

1 : (The local Lipschitz condition) For each integer R , there exists a positive constant K R , such that, for all x , y , x ¯ , y ¯ R d with x y x ¯ y ¯ R ,

f ( x , y ) f ( x ¯ , y ¯ ) 2 g ( x , y ) g ( x ¯ , y ¯ ) 2 K R ( x x ¯ 2 + y y ¯ 2 ) .

2 : There exists a constant k ( 0 , 1 ) such that, for all x , y R d ,

(3) u ( x ) u ( y ) k x y .

Moreover, we suppose that u ( 0 ) = 0 , which, together with (3), implies that

(4) u ( x ) k x .

3 : The delay function δ is continuously differentiable and δ ( t ) δ ¯ < 1 .

4 : There are constants p 2 and K > 0 such that, for all a ( 0 , 1 ] and x , y R d ,

( x a u ( y ) ) T f ( x , y ) + p 1 2 g ( x , y ) 2 K ( 1 + x 2 + a y 2 ) .

Clearly, particularly for a = 1 , we obtain the following condition.

4 : (The Khasminski-type condition) There are constants p 2 and K > 0 such that, for all x , y R d ,

( x u ( y ) ) T f ( x , y ) + p 1 2 g ( x , y ) 2 K ( 1 + x 2 + y 2 ) .

5 : There exists a nonnegative constant C ξ such that, for all p ˆ [ 2 , p ] ,

(5) E sup t , s [ τ , 0 ] , s t Δ ξ ( s ) ξ ( t ) p ˆ C ξ Δ p ˆ 2 .

6 : There exists a positive constant η such that

(6) δ ( t ) δ ( s ) η t s , t , s 0 .

2 Truncated EM approximate solution

Recall the truncated EM numerical scheme defined in [24]. The function μ : R + R + is strictly increasing and continuous, such that μ ( r ) as r and for all r 1 ,

(7) sup x y r ( f ( x , y ) g ( x , y ) u ( y ) ) μ ( r ) .

The inverse function of μ , μ 1 : [ μ ( 1 ) , ) R + is a strictly increasing continuous function. We also choose a constant h ˆ 1 μ ( 1 ) and a strictly decreasing function h : ( 0 , 1 ) [ μ ( 1 ) , ) such that

(8) lim Δ 0 h ( Δ ) = ,

and for all Δ ( 0 , 1 ) , and some ε 0 ( 0 , 1 ) , we have that

(9) Δ 1 ε 0 4 h ( Δ ) h ˆ .

Remark 1

In [24], it was required to choose a number Δ * ( 0 , 1 ) and a strictly decreasing function h : ( 0 , Δ * ] ( 0 , ) such that

(10) h ( Δ * ) μ ( 1 ) , lim Δ 0 h ( Δ ) = ,

and for all Δ ( 0 , Δ * ] and some ε 0 ( 0 , 1 ) , we have that

Δ 1 ε 0 4 h ( Δ ) 1 .

Hence, we simply let Δ * = 1 and remove the condition h ( Δ * ) μ ( 1 ) , while we also replace the condition Δ 1 ε 0 4 h ( Δ ) 1 by a weaker one Δ 1 ε 0 4 h ( Δ ) h ˆ , Δ ( 0 , 1 ) . In other words, we have made the choice of function h more flexible.

In [24], for a given step size Δ ( 0 , 1 ) , a mapping π Δ ( x ) from R d to the closed ball { x R d : x μ 1 ( h ( Δ ) ) } is defined by

(11) π Δ ( x ) = ( x μ 1 ( h ( Δ ) ) ) x x ,

where we set x x = 0 when x = 0 . Then, the truncated functions are defined as

f Δ ( x , y ) = f ( π Δ ( x ) , π Δ ( y ) ) , g Δ ( x , y ) = g ( π Δ ( x ) , π Δ ( y ) ) , and u Δ ( x ) = u ( π Δ ( x ) ) ,

for x , y R d . It is easy to see that

( x μ 1 ( h ( Δ ) ) ) x x μ 1 ( h ( Δ ) ) ,

which together with (7) implies that, for all x , y R d ,

(12) f Δ ( x , y ) g Δ ( x , y ) u Δ ( y ) μ ( μ 1 ( h ( Δ ) ) ) = h ( Δ ) .

So, we consider the following stochastic differential equation:

d [ x ( t ) u Δ ( x ( t δ ( t ) ) ) ] = f Δ ( x ( t ) , x ( t δ ( t ) ) ) d t + g Δ ( x ( t ) , x ( t δ ( t ) ) ) d B ( t ) , t 0 ,

satisfying the initial condition (2).

Moreover, in [24], the truncated EM approximate solution x Δ of equation (1) is defined on the finite time interval [ 0 , T ] , for T > 0 . It is supposed that T τ is a rational number; otherwise, we may replace T by a larger number. Also, it is assumed that the step size Δ ( 0 , 1 ) and Δ = τ M = T N for some integers M > τ and N > T . The discrete-time truncated EM approximate solution X Δ of equation (1) is defined on the equidistant partition t i = i Δ , for i { ( M + 1 ) , M , , 0 , 1 , , N } . In order for this solution to be well defined, it is supposed that

(13) δ ( Δ ) = δ ( 0 ) , X Δ ( ( M + 1 ) Δ ) = ξ ( M Δ ) .

Then, the following definition is imposed:

(14) X Δ ( t i ) = ξ ( t i ) , i = M , ( M 1 ) , , 0 ,

(15) X Δ ( t i + 1 ) = X Δ ( t i ) + u Δ ( X Δ ( t i I Δ [ δ ( t i ) ] Δ ) ) u Δ ( X Δ ( t i 1 I Δ [ δ ( t i 1 ) ] Δ ) ) + f Δ ( X Δ ( t i ) , X Δ ( t i I Δ [ δ ( t i ) ] Δ ) ) Δ + g Δ ( X Δ ( t i ) , X Δ ( t i I Δ [ δ ( t i ) ] Δ ) ) Δ B i , i { 0 , 1 , , N } ,

where Δ B i = B ( t i + 1 ) B ( t i ) and I Δ [ u ] denotes the integer part of the real number u Δ , for u [ 0 , τ ] . For further analysis, it is more convenient to work with the continuous-time approximations. In that sense, the step processes

(16) x ¯ Δ ( t ) = i = 0 N 1 X Δ ( t i ) I [ t i , t i + 1 ) ( t ) , and y ¯ Δ ( t ) = i = 1 N 1 X Δ ( t i I Δ [ δ ( t i ) ] Δ ) I [ t i , t i + 1 ) ( t )

are introduced, where I [ t i , t i + 1 ) ( t ) is the indicator function of [ t i , t i + 1 ) . Then, the linear combination of X Δ ( t i 1 I Δ [ δ ( t i 1 ) ] Δ ) and X Δ ( t i I Δ [ δ ( t i ) ] Δ ) is defined as

(17) Z i ( t ) = y ¯ Δ ( t i 1 ) + t t i Δ ( y ¯ Δ ( t i ) y ¯ Δ ( t i 1 ) ) ,

whenever t [ t i , t i + 1 ) , i = 0 , 1 , 2 , , N 1 . For convenience, the following notation is used:

(18) Z i ( t ) = 1 t t i Δ y ¯ Δ ( t i 1 ) + t t i Δ y ¯ Δ ( t i ) ,

(19) z ¯ Δ ( t ) = i = 0 N 1 Z i ( t ) I [ t i , t i + 1 ) ( t ) .

Then, the continuous-time truncated EM approximate solution { x Δ ( t ) , t [ τ , T ] } is defined, such that x Δ ( t ) = ξ ( t ) , t [ τ , 0 ] , while for t [ t i , t i + 1 ) , i { 0 , 1 , , N 1 } , we have that

(20) x Δ ( t ) = X Δ ( t i ) + u Δ ( Z i ( t ) ) u Δ ( X Δ ( t i 1 I Δ [ δ ( t i 1 ) ] Δ ) ) + t i t f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d s + t i t g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d B ( s ) .

Since, for t [ 0 , T ] , there is a unique integer i 0 such that t [ t i , t i + 1 ) , from (18) and (19), we obtain that for p 2 ,

(21) E z ¯ Δ ( t ) p = E Z i ( t ) p 2 p 1 1 t t i Δ p E y ¯ Δ ( t i 1 ) p + t t i Δ p E y ¯ Δ ( t i ) p 2 p 1 1 t t i Δ sup τ s t E x Δ ( s ) p + t t i Δ sup τ s t E x Δ ( s ) p 2 p 1 sup τ s t E x Δ ( s ) p .

On the other hand, for t [ 0 , T ] , we see that

(22) E y ¯ Δ ( t ) p = E X Δ ( t i I Δ [ δ ( t i ) ] Δ ) p sup τ t i t E X Δ ( t i ) p sup τ s t E x Δ ( s ) p .

Moreover, for all t [ 0 , T ] , we have

(23) E x ¯ Δ ( t ) p sup 0 s t E x ¯ Δ ( s ) p = sup 0 t i t E X Δ ( t i ) p = sup 0 t i t E x Δ ( t i ) p sup 0 s t E x Δ ( s ) p .

In what follows, before stating the main results of this article, we give some assertions from the article [24], which will be used explicitly throughout the article.

Theorem 1

[24, Theorem 1.1] Let Assumptions 1 , 2 , 3 , and 4 be satisfied. Then, for any initial condition ξ C 0 b ( [ τ , 0 ] ; R d ) , there exists a unique global solution x = { x ( t ) , t τ } of equation (1). Moreover, the solution has the property that for all T > 0 , we have that

sup τ t T E x ( t ) p < .

Lemma 1

[24, Lemma 2.1] Let Assumption 2 hold. Then, for k ( 0 , 1 3 ) and all x , y R d ,

(24) u Δ ( x ) u Δ ( y ) 3 k x y .

Moreover, for all x R d ,

(25) u Δ ( x ) k x .

Lemma 2

[24, Lemma 2.2] Let Assumption 4 hold. Then, for every Δ ( 0 , Δ * ] and all x , y R d , we have that

(26) ( x u Δ ( y ) ) T f Δ ( x , y ) + p 1 2 g Δ ( x , y ) 2 K ¯ ( 1 + x 2 + y 2 ) ,

where K ¯ = 3 2 K .

Lemma 3

[24, Lemma 3.1] If Assumptions 2 , 3 , 5 , and 6 hold, together with the condition

(27) k ( 3 + η ) 2 < 1 ,

then, for any Δ ( 0 , Δ * ] and for any integer l > 1 and p 2 ,

(28) E sup M k N X Δ ( t k ) X Δ ( t k 1 ) p c ˜ Δ p 2 + c ˜ l Δ p l 1 2 l ( h ( Δ ) ) p ,

where constants c ˜ > 0 and c ˜ l > 0 are independent of Δ and c ˜ l is dependent on l .

Corollary 1

[24, Corollary 3.2] Let the conditions of Lemma 3 hold. Then, for any integer l > 1 and p 2 ,

(29) E sup Δ t T y ¯ Δ ( t ) z ¯ Δ ( t ) p c ¯ Δ p 2 + c ¯ l Δ p l 1 2 l ( h ( Δ ) ) p ,

(30) E x Δ ( t ) x ¯ Δ ( t ) p c Δ p 2 + c l Δ p l 1 2 l ( h ( Δ ) ) p , t [ 0 , T ] ,

where c ¯ , c , c ¯ l , and c l are positive constants independent of Δ and c ¯ l and c l are dependent on l.

Lemma 4

[24, Lemma 3.4] Let the conditions of Lemma 3 hold together with Assumption 4 . Then, for p > 0 ,

(31) sup 0 < Δ Δ * sup 0 t T E x Δ ( t ) p C ,

where C is a positive real constant dependent on T , p , l , K ¯ , k , and ξ , but independent of Δ .

In the sequel, we will prove that the truncated function u Δ preserves Assumption 2 very well. In that way, we generalize the result of Lemma 1, i.e., instead of the Lipschitz constant k ( 0 , 1 3 ) in that lemma, we will obtain corresponding result with k ( 0 , 1 ) . This constant does not affect assertions in [24], except the one of Lemma 3.

Lemma 5

Let Assumption 2 hold. Then, for k ( 0 , 1 ) and all x , y R d ,

(32) u Δ ( x ) u Δ ( y ) k x y .

Moreover, if u ( 0 ) = 0 , then for all x R d ,

(33) u Δ ( x ) k x .

Proof

Let x , y R d be arbitrary. In the following consideration, we observe x , y R d such that x 0 and y 0 . For x , y R d with x μ 1 ( h ( Δ ) ) and y μ 1 ( h ( Δ ) ) , assertion (32) follows immediately.

On the other hand, for x = ( x 1 , x 2 , , x d ) and y = ( y 1 , y 2 , , y d ) with x > μ 1 ( h ( Δ ) ) and y > μ 1 ( h ( Δ ) ) , on the basis of (3), we obtain that

(34) u Δ ( x ) u Δ ( y ) 2 = u ( π Δ ( x ) ) u ( π Δ ( y ) ) 2 k 2 π Δ ( x ) π Δ ( y ) 2 = k 2 ( μ 1 ( h ( Δ ) ) ) 2 x x y y 2 = k 2 ( μ 1 ( h ( Δ ) ) ) 2 x 1 2 + + x d 2 x 2 2 x 1 y 1 + + x d y d x y + y 1 2 + + y d 2 y 2 .

If x y , we have that

u Δ ( x ) u Δ ( y ) 2 k 2 ( μ 1 ( h ( Δ ) ) ) 2 ( x 1 y 1 ) 2 + + ( x d y d ) 2 x 2 + 2 x 1 y 1 + + x d y d x 2 1 x y y 1 2 + + y d 2 x 2 + 1 = k 2 ( μ 1 ( h ( Δ ) ) ) 2 x y 2 x 2 + 2 x 1 y 1 + + x d y d x 2 1 x y y 2 x 2 + 1 .

Now, on the basis of Cauchy-Schwarz inequality, we obtain

u Δ ( x ) u Δ ( y ) 2 k 2 ( μ 1 ( h ( Δ ) ) ) 2 x y 2 x 2 + 2 x y x 2 1 x y y 2 x 2 + 1 = k 2 ( μ 1 ( h ( Δ ) ) ) 2 x 2 x y 2 k 2 ( μ 1 ( h ( Δ ) ) ) 2 1 y x 2 k 2 ( μ 1 ( h ( Δ ) ) ) 2 x 2 x y 2 .

Using the condition μ 1 ( h ( Δ ) ) x < 1 , we conclude that inequality (32) holds.

On the other hand, for y < x , on the basis of (34), we obtain that

u Δ ( x ) u Δ ( y ) 2 k 2 ( μ 1 ( h ( Δ ) ) ) 2 1 + 2 x 1 y 1 + + x d y d y 2 1 y x x 1 2 + + x d 2 y 2 + ( x 1 y 1 ) 2 + + ( x d y d ) 2 y 2 = k 2 ( μ 1 ( h ( Δ ) ) ) 2 1 + 2 x 1 y 1 + + x d y d y 2 1 y x x 2 y 2 + x y 2 y 2 .

Now, Cauchy-Schwarz inequality yields

u Δ ( x ) u Δ ( y ) 2 k 2 ( μ 1 ( h ( Δ ) ) ) 2 1 + 2 x y y 2 1 y x x 2 y 2 + x y 2 y 2 = k 2 ( μ 1 ( h ( Δ ) ) ) 2 y 2 x y 2 k 2 ( μ 1 ( h ( Δ ) ) ) 2 1 x y 2 < k 2 ( μ 1 ( h ( Δ ) ) ) 2 y 2 x y 2 .

Based on the assumption μ 1 ( h ( Δ ) ) y < 1 , we see that inequality (32) holds.

For x , y R d with x > μ 1 ( h ( Δ ) ) and y μ 1 ( h ( Δ ) ) , bearing in mind (3) and Cauchy-Schwarz inequality, we find that

u Δ ( x ) u Δ ( y ) 2 k 2 π Δ ( x ) π Δ ( y ) 2 = k 2 μ 1 ( h ( Δ ) ) x x 1 y 1 2 + + μ 1 ( h ( Δ ) ) x x d y d 2 = k 2 ( μ 1 ( h ( Δ ) ) ) 2 2 μ 1 ( h ( Δ ) ) x ( x 1 y 1 + + x d y d ) + y 2 = k 2 x y 2 + 2 ( x 1 y 1 + + x d y d ) 1 μ 1 ( h ( Δ ) ) x + μ 1 ( h ( Δ ) ) x 2 x 2 x 2 k 2 x y 2 + 2 x y 1 μ 1 ( h ( Δ ) ) x x 2 1 μ 1 ( h ( Δ ) ) x 1 + μ 1 ( h ( Δ ) ) x = k 2 x y 2 x 2 1 μ 1 ( h ( Δ ) ) x 1 + ( μ 1 ( h ( Δ ) ) ) x 2 y x .

Since x > μ 1 ( h ( Δ ) ) y , then relation (32) holds.

The same way (32) follows for x , y R d with μ 1 ( h ( Δ ) ) x and μ 1 ( h ( Δ ) ) > y .

Now, we suppose that one of the values x and y is equal to null vector, e.g., y = 0 . Then, from (4), we have that

u Δ ( x ) u Δ ( y ) = u ( π Δ ( x ) ) k π Δ ( x ) k x = k x y .

If x = 0 and y = 0 , then (32) holds. It should be pointed out that (33) holds as well.□

Having in mind the previous lemma, we can prove the following one under the condition on the Lipschitz constants of u and δ , which is weaker than the corresponding one from Lemma 3. Since the line of the proof is similar to that of Lemma 3, we present the proof by stressing different parts.

Lemma 6

If Assumptions 2 , 3 , 5 , and 6 hold, together with the condition

(35) k ( 3 + η ) < 1 ,

then, for any Δ ( 0 , 1 ) and for any integer l > 1 and p ˆ [ 2 , p ] ,

E sup M k N X Δ ( t k ) X Δ ( t k 1 ) p ˆ c ˜ Δ p ˆ 2 + c ˜ l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ ,

where constants c ˜ > 0 and c ˜ l > 0 are independent of Δ and c ˜ l is dependent on l .

Proof

Fix any Δ ( 0 , 1 ) . On the basis of (13), (14), and Assumption 5 , we observe that

(36) E sup M k 0 X Δ ( t k ) X Δ ( t k 1 ) p ˆ = E sup M k 0 ξ ( t k ) ξ ( t k 1 ) p ˆ C ξ Δ p ˆ 2 .

When k { 1 , 2 , , N } , from (15), we have

X Δ ( t k ) X Δ ( t k 1 ) = u Δ ( X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) ) u Δ ( X Δ ( t k 2 I Δ [ δ ( t k 2 ) ] Δ ) ) + f Δ ( X Δ ( t k 1 ) , X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) ) Δ + g Δ ( X Δ ( t k 1 ) , X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) ) Δ B k 1 .

Applying the elementary inequality

(37) a + b p ˆ 1 + ε 1 p ˆ 1 p ˆ 1 a p ˆ + b p ˆ ε , a , b R , p ˆ > 1 ,

and on the basis of (32), we obtain

(38) E sup 1 k N X Δ ( t k ) X Δ ( t k 1 ) p ˆ 1 + ε 1 p ˆ 1 p ˆ 1 k p ˆ ε E sup 1 k N X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) X Δ ( t k 2 I Δ [ δ ( t k 2 ) ] Δ ) p ˆ + 2 p ˆ 1 1 + ε 1 p ˆ 1 p ˆ 1 E sup 1 k N f Δ ( X Δ ( t k 1 ) , X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) ) p ˆ Δ p ˆ + 2 p ˆ 1 1 + ε 1 p ˆ 1 p ˆ 1 E sup 1 k N g Δ ( X Δ ( t k 1 ) , X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) ) Δ B k 1 p ˆ .

On the basis of (12), we see that

(39) E sup 1 k N f Δ ( X Δ ( t k 1 ) , X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) ) p ˆ ( h ( Δ ) ) p ˆ .

Now, from the proof of Lemma 3, we have that

(40) E sup 1 k N + 1 X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) X Δ ( t k 2 I Δ [ δ ( t k 2 ) ] Δ ) p ˆ ( 3 + η ) p ˆ E sup M k N X Δ ( t k ) X Δ ( t k 1 ) p ˆ

and

(41) E sup 1 k N g Δ ( X Δ ( t k 1 ) , X Δ ( t k 1 I Δ [ δ ( t k 1 ) ] Δ ) ) Δ B k 1 p ˆ m p ˆ 2 ( T ( 2 p ˆ l 1 ) ! ! ) 1 2 l ( h ( Δ ) ) p ˆ Δ p ˆ l 1 2 l .

Substituting (39), (40), and (41) into (38), bearing in mind that Δ ( 0 , 1 ) , we obtain

(42) E sup 1 k N X Δ ( t k ) X Δ ( t k 1 ) p ˆ 1 + ε 1 p ˆ 1 p ˆ 1 k p ˆ ε ( 3 + η ) p ˆ E sup M k N X Δ ( t k ) X Δ ( t k 1 ) p ˆ + 2 p ˆ 1 1 + ε 1 p ˆ 1 p ˆ 1 1 + m p ˆ 2 ( T ( 2 p ˆ l 1 ) ! ! ) 1 2 l ( h ( Δ ) ) p ˆ Δ p ˆ l 1 2 l .

On the basis of condition (35), there exists ε ¯ > 0 such that

1 + ε ¯ 1 p ˆ 1 p ˆ 1 k p ˆ ε ¯ ( 3 + η ) p ˆ = γ ( 0 , 1 ) .

Applying (36), inequality (42) yields

E sup M k N X Δ ( t k ) X Δ ( t k 1 ) p ˆ C ξ Δ p ˆ 2 + γ E sup M k N X Δ ( t k ) X Δ ( t k 1 ) p ˆ + 2 p ˆ 1 1 + ε ¯ 1 p ˆ 1 p ˆ 1 1 + m p ˆ 2 ( T ( 2 p ˆ l 1 ) ! ! ) 1 2 l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ .

Thus, we find that

E sup M k N X Δ ( t k ) X Δ ( t k 1 ) p ˆ c ˜ Δ p ˆ 2 + c ˜ l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ ,

where

c ˜ = C ξ 1 γ , c ˜ l = 2 p ˆ 1 1 γ 1 + ε ¯ 1 p ˆ 1 p ˆ 1 1 + m p ˆ 2 ( T ( 2 p ˆ l 1 ) ! ! ) 1 2 l ,

which completes the proof.□

The following lemma shows that the truncated functions f Δ and g Δ preserve Assumption 4 very well and that inequality will be applied explicitly in the proofs of the main results of this article. Some parts of the proof will be omitted since they are the same as in the proof of Lemma 2.

Lemma 7

Let Assumption 4 hold. Then, for every Δ ( 0 , 1 ) and x , y R d , we have

(43) ( x u Δ ( y ) ) T f Δ ( x , y ) + p 1 2 g Δ ( x , y ) 2 K ˆ ( 1 + x 2 + y 2 ) ,

where K ˆ = 3 2 K 1 1 μ 1 ( h ( Δ ) ) .

Proof

This lemma essentially represents Lemma 2, but here we show that the condition h ( Δ * ) μ ( 1 ) is not necessary.

Fix any Δ ( 0 , 1 ) . For x , y R d with x y < μ 1 ( h ( Δ ) ) , on the basis of Assumption 4 , inequality (43) obviously holds for a = 1 .

Furthermore, if x y μ 1 ( h ( Δ ) ) or x μ 1 ( h ( Δ ) ) and y < μ 1 ( h ( Δ ) ) , then the proof of Lemma 2 shows that

( x u Δ ( y ) ) T f Δ ( x , y ) + p 1 2 g Δ ( x , y ) 2 K x μ 1 ( h ( Δ ) ) + x 2 + y 2 .

On the basis of μ 1 ( h ( Δ ) ) μ 1 ( h ( 1 ) ) , we obtain

( x u Δ ( y ) ) T f Δ ( x , y ) + p 1 2 g Δ ( x , y ) 2 K x μ 1 ( h ( 1 ) ) + x 2 + y 2 K 1 1 μ 1 ( h ( Δ ) ) [ x + x 2 + y 2 ] K ˆ [ 1 + x 2 + y 2 ] ,

where K ˆ = 3 2 K 1 1 μ 1 ( h ( Δ ) ) . Finally, on the basis of the proof of Lemma 2, we know that assertion (43) holds for y μ 1 ( h ( Δ ) ) and x < μ 1 ( h ( Δ ) ) , which completes the proof.□

In view of Lemma 7, we see that the constant K ¯ in Lemma 2 is now replaced by another constant K ˆ . Based on the fact that condition h ( Δ * ) μ ( 1 ) was only used in [24] to prove Lemma 2, we can conclude that it does not affect any other result in [24]. Also, it is easy to check that replacing Δ 1 ε 0 4 h ( Δ ) 1 by Δ 1 ε 0 4 h ( Δ ) h ˆ does not make any effect on the other results in [24], so they can be applied in further analysis in this article.

3 Convergence rates

In [24], the L q -convergence of the truncated EM method is established for q [ 2 , p ) , where p is a parameter in Assumption 4 . However, the convergence was in the asymptotic form without the convergence rate. In this section, we will determine the convergence rate of the method without restrictions on the functions under consideration, which sometimes force the step size to be so small that the application of the method would be limited.

For further consideration, the following hypotheses are essential.

7 : There exists a pair of constants q > 2 and H 1 > 0 such that, for all x , y , x ¯ , y ¯ R d ,

(44) ( x u ( y ) x ¯ + u ( y ¯ ) ) T ( f ( x , y ) f ( x ¯ , y ¯ ) ) + q 1 2 g ( x , y ) g ( x ¯ , y ¯ ) 2 H 1 ( x x ¯ 2 + y y ¯ 2 ) .

8 : There exists a pair of positive constants ρ and H 2 such that, for all x , y , x ¯ , y ¯ R d ,

(45) f ( x , y ) f ( x ¯ , y ¯ ) 2 g ( x , y ) g ( x ¯ , y ¯ ) 2 H 2 ( 1 + x ρ + y ρ + x ¯ ρ + y ¯ ρ ) ( x x ¯ 2 + y y ¯ 2 ) .

It is useful to emphasize that the truncated functions f Δ and g Δ preserve Assumption 8 perfectly, which is essential for proving the theorem determining the rate of the convergence of the truncated EM method. In fact, we derive that, for all x , y , x ¯ , y ¯ R d ,

(46) f Δ ( x , y ) f Δ ( x ¯ , y ¯ ) 2 g Δ ( x , y ) g Δ ( x ¯ , y ¯ ) 2 = f ( π Δ ( x ) , π Δ ( y ) ) f ( π Δ ( x ¯ ) , π Δ ( y ¯ ) ) 2 g ( π Δ ( x ) , π Δ ( y ) ) g ( π Δ ( x ¯ ) , π Δ ( y ¯ ) ) 2 H 2 ( 1 + π Δ ( x ) ρ + π Δ ( y ) ρ + π Δ ( x ¯ ) ρ + π Δ ( y ¯ ) ρ ) ( π Δ ( x ) π Δ ( x ¯ ) 2 + π Δ ( y ) π Δ ( y ¯ ) 2 ) .

Based on the definition of the function π Δ ( ) , we see that, for all x R d ,

(47) π Δ ( x ) x .

Moreover, we observe from Lemma 5 that, for all x , y R d ,

π Δ ( x ) π Δ ( y ) 2 x y 2 .

Therefore, (46) can be estimated as

(48) f Δ ( x , y ) f Δ ( x ¯ , y ¯ ) 2 g Δ ( x , y ) g Δ ( x ¯ , y ¯ ) 2 H 2 ( 1 + x ρ + y ρ + x ¯ ρ + y ¯ ρ ) ( x x ¯ 2 + y y ¯ 2 ) .

On the other hand, from Assumption 8 , we have that, for all x , y 1 ,

f ( x , y ) f ( 0 , 0 ) 2 f ( x , y ) f ( 0 , 0 ) 2 H 2 ( 1 + x ρ + y ρ ) ( x 2 + y 2 ) 2 H 2 ( x ρ + y ρ ) ( x 2 + y 2 ) 8 H 2 ( x y ) ρ + 2 .

In a similar way, we can estimate the expression g ( x , y ) g ( 0 , 0 ) 2 . So, we obtain that, for all x , y 1 ,

(49) f ( x , y ) g ( x , y ) H 3 ( x y ) ρ + 2 2 ,

where H 3 = ( 8 H 2 + f ( 0 , 0 ) + g ( 0 , 0 ) ) . Namely, both coefficients f and g grow at most polynomially, whence we can let μ ( u ) = H 3 u ρ + 2 2 . Moreover, in view of (9), we can let h ( Δ ) = h ˆ Δ 1 ε 4 for some ε [ ε 0 , 1 ) . In other words, there are lots of choices for μ ( ) and h ( ) .

For completing this section, we establish the following two lemmas that are essential for obtaining the main results of this article. In that sense, we first impose the following remark.

Remark 2

It should be emphasized that, when applying Lemmas 3 and 4 in the proofs of the following assertions, based on Remark 1, we will consider that Δ ( 0 , 1 ) instead of Δ ( 0 , Δ * ] .

In the next lemma, we will estimate the delayed terms of the continuous x Δ ( t δ ( t ) ) and discrete z ¯ Δ ( t ) truncated EM solutions.

Lemma 8

Let Assumptions 2 , 3 , 5 , and 6 hold together with condition (35). Then, for any Δ ( 0 , 1 ) and any integers l > 1 and p ˆ [ 2 , p ] ,

E x Δ ( t δ ( t ) ) z ¯ Δ ( t ) p ˆ c ˇ Δ p ˆ 2 + c ˇ l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ , t [ 0 , T ] ,

where c ˇ and c ˇ l are positive constants independent of Δ and c ˇ l is dependent on l .

Proof

Fix any t [ 0 , T ] . There is unique i { 0 , 1 , 2 , , N 1 } such that t [ t i , t i + 1 ) , and let j { M , ( M 1 ) , , i } be such that t δ ( t ) [ t j , t j + 1 ) . Observe that, for t i 1 I Δ [ δ ( t i 1 ) ] Δ t j t δ ( t ) , in a view of 6 , we have that

t j t i 1 + I Δ [ δ ( t i 1 ) ] Δ t δ ( t ) t i 1 + I Δ [ δ ( t i 1 ) ] Δ 2 Δ + I Δ [ δ ( t i 1 ) ] Δ δ ( t ) 3 Δ + δ ( t i 1 ) δ ( t ) ( 4 + 2 η ) Δ .

Otherwise, for t j t δ ( t ) t i 1 I Δ [ δ ( t i 1 ) ] Δ , we obtain

t j t i 1 + I Δ [ δ ( t i 1 ) ] Δ Δ + t i 1 I Δ [ δ ( t i 1 ) ] Δ t + δ ( t ) 4 Δ + δ ( t ) δ ( t i 1 ) ( 5 + 2 η ) Δ .

Therefore, we obtain

(50) j ( i 1 ) I Δ [ δ ( t i 1 ) ] 5 + 2 η .

On the other hand, on the basis of 6 , we have that, for α = 2 , 3 , , N + 1 ,

t α 1 I Δ [ δ ( t α 1 ) ] Δ t α 2 + I Δ [ δ ( t α 2 ) ] Δ Δ + I Δ [ δ ( t α 2 ) ] I Δ [ δ ( t α 1 ) ] Δ Δ + ( 1 + η ) Δ ( 3 + η ) Δ .

For α = 1 , since δ ( Δ ) = δ ( 0 ) , we find that t α 1 I Δ [ δ ( t α 1 ) ] Δ t α 2 + I Δ [ δ ( t α 2 ) ] Δ = Δ . So, for α = 1 , 2 , , N + 1 , we obtain

(51) ( α 1 ) I Δ [ δ ( t α 1 ) ] ( α 2 ) + I Δ [ δ ( t α 2 ) ] ( 3 + η ) .

Now, by Definition (19) of the step-process z ¯ Δ ( t ) , we see that for t [ t i , t i + 1 ) ,

(52) E x Δ ( t δ ( t ) ) z ¯ ( t ) p ˆ 2 p ˆ 1 E x Δ ( t δ ( t ) ) X Δ ( t j ) p ˆ + 2 p ˆ 1 E X Δ ( t j ) Z i ( t ) p ˆ .

In order to estimate E x Δ ( t δ ( t ) ) X Δ ( t j ) p ˆ , we discuss the following two cases.

Case 1: If j 1 , then, from Assumption 5 , we obtain that

E x Δ ( t δ ( t ) ) X Δ ( t j ) p ˆ E sup t [ 0 , T ] , j 1 X Δ ( t δ ( t ) ) X Δ ( t j ) p ˆ E sup s , t [ τ , 0 ] , s t Δ ξ ( t ) ξ ( s ) p ˆ C ξ Δ p ˆ 2 .

Case 2: If j 0 , then, on the basis of (16), (17), (20), and Lemma 5, we have

(53) E x Δ ( t δ ( t ) ) X Δ ( t j ) p ˆ 3 p ˆ 1 E u Δ ( Z j ( t δ ( t ) ) ) u Δ ( X Δ ( t j 1 I Δ [ δ ( t j 1 ) ] Δ ) ) p ˆ + 3 p ˆ 1 E t j t δ ( t ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d s p ˆ + 3 p ˆ 1 E t j t δ ( t ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d B ( s ) p ˆ 3 p ˆ 1 k p ˆ E Z j ( t δ ( t ) ) X Δ ( t j 1 I Δ [ δ ( t j 1 ) ] Δ ) p ˆ + 3 p ˆ 1 E t j t δ ( t ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d s p ˆ + 3 p ˆ 1 E t j t δ ( t ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d B ( s ) p ˆ = 3 2 p ˆ 1 k p ˆ t δ ( t ) t j Δ p ˆ E X Δ ( t j I Δ [ δ ( t j ) ] Δ ) X Δ ( t j 1 I Δ [ δ ( t j 1 ) ] Δ ) p ˆ + 3 p ˆ 1 E t j t δ ( t ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d s p ˆ + 3 p ˆ 1 E t j t δ ( t ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d B ( s ) p ˆ .

Now, we estimate the expression E X Δ ( t j I Δ [ δ ( t j ) ] Δ ) X Δ ( t j 1 I Δ [ δ ( t j 1 ) ] Δ ) p ˆ . Since, on the basis of 3 , we have that ( j 1 ) I Δ [ δ ( t j 1 ) ] ( j 2 ) I Δ [ δ ( t j 2 ) ] for any j { 1 , 2 , , N + 1 } , using (51), we observe that

(54) X Δ ( t j 1 I Δ [ δ ( t j 1 ) ] Δ ) X Δ ( t j 2 I Δ [ δ ( t j 2 ) ] Δ ) p ˆ ( 3 + η ) p ˆ 1 α = ( j 2 ) I Δ [ δ ( t j 2 ) ] + 1 ( j 1 ) I Δ [ δ ( t j 1 ) ] X Δ ( t α ) X Δ ( t α 1 ) p ˆ ( 3 + η ) p ˆ sup M α N X Δ ( t α ) X Δ ( t α 1 ) p ˆ .

In a similar way, using (50), we obtain

(55) X Δ ( t j ) X Δ ( t i 1 I Δ [ δ ( t i 1 ) ] Δ ) p ˆ ( 5 + 2 η ) p ˆ 1 α = ( i 1 ) I Δ [ δ ( t i 1 ) ] + 1 j X Δ ( t α ) X Δ ( t α 1 ) p ˆ ( 5 + 2 η ) p ˆ sup M α N X Δ ( t α ) X Δ ( t α 1 ) p ˆ .

Then, on the basis of Lemma 6, we conclude that

(56) E X Δ ( t j I Δ [ δ ( t j ) ] Δ ) X Δ ( t j 1 I Δ [ δ ( t j 1 ) ] Δ ) p ˆ ( 3 + η ) p ˆ 1 c ˜ Δ p ˆ 2 + c ˜ l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ .

By the Hölder and Burkholder-Davis-Gundy inequalities, as well as (12), we obtain

(57) E t j t δ ( t ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d s p ˆ + E t j t δ ( t ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) d B ( s ) p ˆ ( t δ ( t ) t j ) p ˆ 1 t j t δ ( t ) E f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) p ˆ d s + p ˆ ( p ˆ 1 ) 2 p ˆ 2 ( t δ ( t ) t j ) p ˆ 2 1 t j t δ ( t ) E g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) p ˆ d s Δ p ˆ ( h ( Δ ) ) p ˆ + p ˆ ( p ˆ 1 ) 2 p ˆ 2 Δ p ˆ 2 ( h ( Δ ) ) p ˆ .

Substituting (56) and (57) into (53), we obtain that

(58) E x Δ ( t δ ( t ) ) X Δ ( t j ) p ˆ 3 p ˆ 1 k p ˆ ( 3 + η ) p ˆ c ˜ Δ p ˆ 2 + c ˜ l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ + 3 p ˆ 1 Δ p ˆ ( h ( Δ ) ) p ˆ + p ˆ ( p ˆ 1 ) 2 p ˆ 2 Δ p ˆ 2 ( h ( Δ ) ) p ˆ .

In order to estimate the second summand of (52), we use Definition (17) and obtain that

(59) X Δ ( t j ) Z k ( t ) = X Δ ( t j ) y ¯ Δ ( t i 1 ) t t i Δ ( y ¯ Δ ( t i ) y ¯ Δ ( t i 1 ) ) X Δ ( t j ) X Δ ( t i 1 I Δ [ δ ( t i 1 ) ] Δ ) t t i Δ ( X Δ ( t i I Δ [ δ ( t i ) ] Δ ) X Δ ( t i 1 I Δ [ δ ( t i 1 ) ] Δ ) ) ,

whenever t [ t i , t i + 1 ) . Based on (54) and (55), we have that

(60) X Δ ( t j ) Z i ( t ) p ˆ 2 p ˆ 1 X Δ ( t j ) X Δ ( t i 1 I Δ [ δ ( t i 1 ) ] Δ ) p ˆ + 2 p ˆ 1 X Δ ( t i I Δ [ δ ( t i ) ] Δ ) X Δ ( t i 1 I Δ [ δ ( t i 1 ) ] Δ ) p ˆ 2 p ˆ 1 ( ( 5 + 2 η ) p ˆ + ( 3 + η ) p ˆ ) sup M α N X Δ ( t α ) X Δ ( t α 1 ) p ˆ .

Then, in a view of Lemma 6, we obtain

(61) E X Δ ( t j ) Z i ( t ) p ˆ 2 p ˆ 1 ( ( 5 + 2 η ) p ˆ + ( 3 + η ) p ˆ ) ( c ˜ Δ p ˆ 2 + c ˜ l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ ) .

Therefore, since Δ ( 0 , 1 ) , substituting (58) and (61) into (52) gives

E x Δ ( t δ ( t ) ) z ¯ ( t ) p ˆ 2 p ˆ 1 ( 3 p ˆ 1 k p ˆ ( 3 + η ) p ˆ + 2 p ˆ 1 ( ( 5 + 2 η ) p ˆ + ( 3 + η ) p ˆ ) ) c ˜ Δ p ˆ 2 + c ˜ l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ + 2 p ˆ 1 3 p ˆ 1 Δ p ˆ ( h ( Δ ) ) p ˆ + p ˆ ( p ˆ 1 ) 2 p ˆ 2 Δ p ˆ 2 ( h ( Δ ) ) p ˆ c ˇ Δ p ˆ 2 + c ˇ l Δ p ˆ l 1 2 l ( h ( Δ ) ) p ˆ ,

where

c ˇ = 2 p ˆ 1 c ˜ ( 3 p ˆ 1 k p ˆ ( 3 + η ) p ˆ + 2 p ˆ 1 ( ( 5 + 2 η ) p ˆ + ( 3 + η ) p ˆ ) ) , c ˇ l = 2 p ˆ 1 c ˜ l ( 3 p ˆ 1 k p ˆ ( 3 + η ) p ˆ + 2 p ˆ 1 ( ( 5 + 2 η ) p ˆ + ( 3 + η ) p ˆ ) ) + 6 p ˆ 1 1 + p ˆ ( p ˆ 1 ) 2 p ˆ 2 .

Now, we are in a position to estimate the closeness between the delayed term of the continuous truncated EM solution x Δ ( t δ ( t ) ) and the truncated delayed term of the discrete truncated EM solution π Δ ( z ¯ Δ ( t ) ) .

Corollary 2

Let the conditions of Lemma 8 hold, together with Assumption 4 . Then, for any integer l > 1 , q ˜ [ 2 , p ) , and Δ ( 0 , 1 ) ,

E x Δ ( t δ ( t ) ) π Δ ( z ¯ Δ ( t ) ) q ˜ s ˇ Δ q ˜ 2 + s ˇ ( μ 1 ( h ( Δ ) ) ) ( p q ˜ ) + s ˇ l Δ q ˜ l 1 2 l ( h ( Δ ) ) q ˜ , t [ 0 , T ] ,

where s ˇ and s ˇ l are positive constants independent of Δ and s ˇ l is dependent on l .

Proof

Fix any t [ 0 , T ] . By Lemma 8, we have that

(62) E x Δ ( t δ ( t ) ) π Δ ( z ¯ Δ ( t ) ) q ˜ 2 q ˜ 1 E x Δ ( t δ ( t ) ) z ¯ Δ ( t ) q ˜ + 2 q ˜ 1 E z ¯ Δ ( t ) π Δ ( z ¯ Δ ( t ) ) q ˜ 2 q ˜ 1 c ˇ Δ q ˜ 2 + c ˇ l Δ q ˜ l 1 2 l ( h ( Δ ) ) q ˜ + E z ¯ Δ ( t ) π Δ ( z ¯ Δ ( t ) ) q ˜ .

On the other hand, by the Hölder and Chebyshev inequalities, we obtain

E z ¯ Δ ( t ) π Δ ( z ¯ Δ ( t ) ) q ˜ = E ( I { z ¯ Δ ( t ) > μ 1 ( h ( Δ ) ) } z ¯ Δ ( t ) q ˜ ) ( P { z ¯ Δ ( t ) > μ 1 ( h ( Δ ) ) } ) p q ˜ p ( E z ¯ Δ ( t ) p ) q ˜ p E z ¯ Δ ( t ) p ( μ 1 ( h ( Δ ) ) ) p p q ˜ p ( E z ¯ Δ ( t ) p ) q ˜ p ( μ 1 ( h ( Δ ) ) ) ( p q ˜ ) E z ¯ Δ ( t ) p .

So, in a view of (21) and Lemma 6, we find that

(63) E z ¯ Δ ( t ) π Δ ( z ¯ Δ ( t ) ) q ˜ 2 p 1 ( μ 1 ( h ( Δ ) ) ) ( p q ˜ ) sup τ s t E x Δ ( s ) p 2 p 1 ( sup τ s 0 E ξ ( s ) p + C ) ( μ 1 ( h ( Δ ) ) ) ( p q ˜ ) .

Thus, substitution of (63) into (62) implies that

E x ( t δ ( t ) ) π Δ ( z ¯ Δ ( t ) ) q ˜ 2 q 1 c ˇ Δ q ˜ 2 + c ˇ l Δ q ˜ l 1 2 l ( h ( Δ ) ) q ˜ + 2 p + q ˜ 2 ( μ 1 ( h ( Δ ) ) ) ( p q ˜ ) ( sup τ s 0 E ξ ( s ) p + C ) s ˇ Δ q ˜ 2 + s ˇ ( μ 1 ( h ( Δ ) ) ) ( p q ˜ ) + s ˇ l Δ q ˜ l 1 2 l ( h ( Δ ) ) q ˜ ,

where

s ˇ = 2 q ˜ 1 ( c ˇ 2 p 1 ( sup τ s 0 E ξ ( s ) p + C ) ) , s ˇ l = 2 q ˜ 1 c ˇ l .

Therefore, the proof is complete.□

In what follows, we will prove the main result of this article which establishes the rate of convergence of the continuous truncated EM solution x Δ ( t ) to the exact solution x ( t ) . Moreover, it establishes the convergence rate of the discrete truncated EM solution x ¯ Δ ( t ) to the exact solution, which is more important from the practical point in view.

Theorem 2

Let Assumptions 1 8 and 4 hold, together with conditions (35) and 2 p > ( 2 + ρ ) q . Then, for any q ¯ [ 2 , q ) , Δ ( 0 , 1 ) and l > 1 ,

(64) sup 0 t T E x Δ ( t ) x ( t ) q ¯ D ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + D Δ q ¯ 2 + D l Δ q ¯ 2 1 2 l ( h ( Δ ) ) q ¯ ,

(65) sup 0 t T E x ¯ Δ ( t ) x ( t ) q ¯ D ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + D Δ q ¯ 2 + D l Δ q ¯ 2 1 2 l ( h ( Δ ) ) q ¯ ,

where D , D , D l , and D l are positive constants independent of Δ and D l and D l are dependent on l .

In particular, recalling (49), we may define, for u 1 ,

(66) μ ( u ) = H 3 u ρ + 2 2 ,

and let h ( Δ ) = h ˆ Δ ε for some ε 0 , 1 ε 0 4 and h ˆ 1 , to obtain

(67) sup 0 t T E x ( t ) x Δ ( t ) q ¯ = O Δ ε 2 p ρ + 2 q ¯ q ¯ 2 1 2 l ε q ¯ ,

(68) sup 0 t T E x ( t ) x ¯ Δ ( t ) q ¯ = O Δ ε 2 p ρ + 2 q ¯ q ¯ 2 1 2 l ε q ¯ .

Proof

First, we stress that Assumption 1 will not be used explicitly in the proof, but its role is important for the existence and uniqueness of the exact solution x .

Fix q ¯ [ 2 , q ) and Δ ( 0 , 1 ) arbitrarily. For t [ 0 , T ] , let us denote

e Δ ( t ) x ( t ) x Δ ( t ) , e ¯ Δ ( t ) x ( t ) u ( x ( t δ ( t ) ) ) x Δ ( t ) + u Δ ( z ¯ Δ ( t ) ) .

For each integer n > x 0 = ξ , we define the stopping time

θ n = inf { t 0 : x ( t ) x Δ ( t ) n } ,

where inf = . By the Ito formula, we have that, for any t [ 0 , T ] ,

(69) E e ¯ Δ ( t θ n ) q ¯ = E 0 t θ n q ¯ e ¯ Δ ( s ) q ¯ 2 ( ( e ¯ Δ ( s ) ) T ( f ( x ( s ) , x ( s δ ( s ) ) ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) ) + q ¯ 1 2 g ( x ( s ) , x ( t δ ( s ) ) ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) 2 d s .

Note that

q ¯ 1 2 g ( x ( s ) , x ( t δ ( s ) ) ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) 2 q ¯ 1 2 1 + q q ¯ q ¯ 1 g ( x ( s ) , x ( t δ ( s ) ) ) g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) 2 + 1 + q ¯ 1 q q ¯ g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) 2 q 1 2 g ( x ( s ) , x ( t δ ( s ) ) ) g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) 2 + ( q ¯ 1 ) ( q 1 ) 2 ( q q ¯ ) g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) 2 .

Then, (69) can be estimated as

(70) E e ¯ Δ ( t θ n ) q ¯ E 0 t θ n q ¯ e ¯ Δ ( s ) q ¯ 2 ( e ¯ Δ ( s ) ) T ( f ( x ( s ) , x ( s δ ( s ) ) ) f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) ) + q 1 2 g ( x ( s ) , x ( s δ ( s ) ) ) g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) 2 + ( e ¯ Δ ( s ) ) T ( f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) ) + ( q ¯ 1 ) ( q 1 ) 2 ( q q ¯ ) g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) 2 d s = J 1 + J 2 ,

where

(71) J 1 = E 0 t θ n q ¯ e ¯ Δ ( s ) q ¯ 2 ( e ¯ Δ ( s ) ) T ( f ( x ( s ) , x ( s δ ( s ) ) f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) ) + q 1 2 g ( x ( s ) , x ( s δ ( s ) ) ) g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) 2 d s , J 2 = E 0 t θ n q ¯ e ¯ Δ ( s ) q ¯ 2 ( e ¯ Δ ( s ) ) T ( f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) ) + ( q ¯ 1 ) ( q 1 ) 2 ( q q ¯ ) g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) 2 d s .

First, we estimate the integral J 1 . Let a , b 0 . Using the elementary inequality

(72) a m ¯ b 1 m ¯ m ¯ a + ( 1 m ¯ ) b , m ¯ [ 0 , 1 ] ,

we derive that

(73) a q ¯ 2 b 2 = ( a q ¯ ) q ¯ 2 q ¯ ( b q ¯ ) 2 q ¯ q ¯ 2 q ¯ a q ¯ + 2 q ¯ b q ¯ .

Furthermore, in a view of Assumption 7 , (73), and relation 0 t θ n t T , we find that

(74) J 1 q ¯ H 1 E 0 t θ n e ¯ Δ ( s ) q ¯ 2 x ( s ) x Δ ( s ) 2 d s + q ¯ H 1 E 0 t θ n e ¯ Δ ( s ) q ¯ 2 x ( s δ ( s ) ) π Δ ( z ¯ Δ ( s ) ) 2 d s q ¯ H 1 E 0 t θ n q ¯ 2 q ¯ e ¯ Δ ( s ) q ¯ + 2 q ¯ x ( s ) x Δ ( s ) q ¯ d s + q ¯ H 1 E 0 t θ n q ¯ 2 q ¯ e ¯ Δ ( s ) q ¯ + 2 q ¯ x ( s δ ( s ) ) π Δ ( z ¯ Δ ( s ) ) q ¯ d s = 2 H 1 ( q ¯ 2 ) E 0 t θ n e ¯ Δ ( s ) q ¯ d s + 2 H 1 E 0 t θ n ( x ( s ) x Δ ( s ) q ¯ + x ( s δ ( s ) ) π Δ ( z ¯ Δ ( s ) ) q ¯ ) d s 2 H 1 ( q ¯ 2 ) E 0 t θ n e ¯ Δ ( s ) q ¯ d s + 2 q ¯ H 1 E 0 t θ n x ( s ) x Δ ( s ) q ¯ d s + E 0 t θ n x ( s δ ( s ) ) x Δ ( s δ ( s ) ) q ¯ d s + E 0 t θ n x Δ ( s δ ( s ) ) π Δ ( z ¯ Δ ( s ) ) q ¯ d s .

Using Assumption 3 , as well as the fact that solutions x and x Δ satisfy the same initial condition, we obtain

2 q ¯ H 1 E 0 t θ n x ( s δ ( s ) ) x Δ ( s δ ( s ) ) q ¯ d s 2 q ¯ H 1 1 δ ¯ E 0 t θ n x ( s ) x Δ ( s ) q ¯ d s .

Substituting the previous estimate into (74), we obtain that

J 1 2 H 1 ( q ¯ 2 ) E 0 t θ n e ¯ Δ ( s ) q ¯ d s + 2 q ¯ 1 + 1 1 δ ¯ H 1 E 0 t θ n x ( s ) x Δ ( s ) q ¯ d s + 2 q ¯ H 1 0 T E x Δ ( s δ ( s ) ) π Δ ( z ¯ Δ ( s ) ) q ¯ d s .

So, applying Corollary 2, we can conclude that for l > 1 ,

(75) J 1 2 H 1 ( q ¯ 2 ) E 0 t θ n e ¯ Δ ( s ) q ¯ d s + 2 q ¯ 1 + 1 1 δ ¯ H 1 E 0 t θ n e Δ ( s ) q ¯ d s + 2 2 q ¯ 1 H 1 T s ˇ Δ q ¯ 2 + s ˇ ( μ 1 ( h ( Δ ) ) ) ( p q ¯ ) + s ˇ l Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ .

The function μ 1 : [ μ ( 1 ) , ) R + is strictly increasing and continuous, such that

(76) ( μ 1 ( h ( Δ ) ) ) ( p q ¯ ) ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 .

Therefore, (75) can be estimated as

(77) J 1 2 H 1 ( q ¯ 2 ) E 0 t θ n e ¯ Δ ( s ) q ¯ d s + 2 q ¯ 1 + 1 1 δ ¯ H 1 E 0 t θ n e Δ ( s ) q ¯ d s + 2 2 q ¯ 1 H 1 T s ˇ Δ q ¯ 2 + s ˇ ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + s ˇ l Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ .

On the other hand, the integral J 2 , given by (71), can be estimated as

(78) J 2 J 21 + J 22 ,

where

J 21 = E 0 t θ n q ¯ e ¯ Δ ( s ) q ¯ 2 ( e ¯ Δ ( s ) ) T ( f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) f Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) ) + ( q ¯ 1 ) ( q 1 ) q q ¯ g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) 2 d s , J 22 = E 0 t θ n q ¯ e ¯ Δ ( s ) q ¯ 2 ( e ¯ Δ ( s ) ) T ( f Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) ) + ( q ¯ 1 ) ( q 1 ) q q ¯ g Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) 2 d s .

In order to estimate J 21 , we use (73) and relation 0 t θ n t T . So, we obtain that

(79) J 21 q ¯ 2 E 0 t θ n e ¯ Δ ( s ) q ¯ d s + q ¯ 2 E 0 t θ n e ¯ Δ ( s ) q ¯ 2 f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) f Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) 2 d s + q ¯ ( q ¯ 1 ) ( q 1 ) q q ¯ E 0 t θ n e ¯ Δ ( s ) q ¯ 2 g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) 2 d s q ¯ 2 E 0 t θ n e ¯ Δ ( s ) q ¯ d s + q ¯ 2 E 0 t θ n q ¯ 2 q ¯ e ¯ Δ ( s ) q ¯ + 2 q ¯ f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) f Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) q ¯ d s + q ¯ ( q ¯ 1 ) ( q 1 ) q q ¯ E 0 t θ n q ¯ 2 q ¯ e ¯ Δ ( s ) q ¯ + 2 q ¯ g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) q ¯ d s q ¯ 1 + ( q ¯ 1 ) ( q 1 ) ( q ¯ 2 ) q q ¯ E 0 t θ n e ¯ Δ ( s ) q ¯ d s + E 0 t θ n f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) f Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) q ¯ d s + 2 ( q ¯ 1 ) ( q 1 ) q q ¯ E 0 t θ n g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) q ¯ d s C 1 E 0 t θ n e ¯ Δ ( s ) q ¯ d s + J 23 ,

where

J 23 = C 1 E 0 t θ n ( f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) f Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) q ¯ + g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) q ¯ ) d s , C 1 = max q ¯ 1 + ( q ¯ 1 ) ( q 1 ) ( q ¯ 2 ) q q ¯ , 1 , 2 ( q ¯ 1 ) ( q 1 ) q q ¯ .

Based on Assumption 8 and (47), the integral J 23 can be estimated as

J 23 C 1 E 0 T ( f ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) f ( π Δ ( x Δ ( s ) ) , π Δ ( z ¯ Δ ( s ) ) ) q ¯ + g ( x Δ ( s ) , π Δ ( z ¯ Δ ( s ) ) ) g ( π Δ ( x Δ ( s ) ) , π Δ ( z ¯ Δ ( s ) ) ) q ¯ ) d s 2 C 1 E 0 T ( H 2 ( 1 + x Δ ( s ) ρ + π Δ ( x Δ ( s ) ) ρ + 2 π Δ ( z ¯ Δ ( s ) ) ρ ) × x Δ ( s ) π Δ ( x Δ ( s ) ) 2 ) q ¯ 2 d s 2 C 1 H 2 q ¯ 2 E 0 T ( 1 + 2 x Δ ( s ) ρ + 2 z ¯ Δ ( s ) ρ ) q ¯ 2 ( x Δ ( s ) π Δ ( x Δ ( s ) ) 2 ) q ¯ 2 d s 2 q ¯ 2 + 1 3 q ¯ 2 1 C 1 H 2 q ¯ 2 0 T E 1 + x Δ ( s ) q ¯ ρ 2 + z ¯ Δ ( s ) q ¯ ρ 2 x Δ ( s ) π Δ ( x Δ ( s ) ) q ¯ d s .

Using the Hölder inequality, together with (21), we observe that

(80) J 23 2 q ¯ 2 + 1 3 q ¯ 2 1 C 1 H 2 q ¯ 2 0 T E 1 + x Δ ( s ) q ¯ ρ 2 + z ¯ Δ ( s ) q ¯ ρ 2 2 p q ¯ ρ q ¯ ρ 2 p × E x Δ ( s ) π Δ ( x Δ ( s ) ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p d s 2 q ¯ 2 + 1 3 q ¯ ( p ρ ) 2 p C 1 H 2 q ¯ 2 0 T [ 1 + ( 1 + 2 p ) sup τ r s E x Δ ( r ) p ] q ¯ ρ 2 p E x Δ ( s ) π Δ ( x Δ ( s ) ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p d s .

Applying the similar procedure as in (63), we obtain that

E x Δ ( s ) π Δ ( x Δ ( s ) ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p = E I { x Δ ( s ) > μ 1 ( h ( Δ ) ) } x Δ ( s ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p [ P { x Δ ( s ) > μ 1 ( h ( Δ ) ) } ] 2 p ( 2 + ρ ) q ¯ 2 p [ E x Δ ( s ) p ] q ¯ p ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 [ E x Δ ( s ) p ] 2 p ρ q ¯ 2 p

Therefore, based on Lemma 4, we estimate (80) as

(81) J 23 2 q ¯ 2 + 1 3 q ¯ ( p ρ ) 2 p C 1 ( 1 + ( 1 + 2 p 1 ) ( sup τ s 0 E ξ ( s ) p + C ) ) q ¯ ρ 2 p H 2 q ¯ 2 × 0 T ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 [ E x Δ ( s ) p ] 2 p ρ q ¯ 2 p d s C ˜ 1 ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 ,

where

C ˜ 1 = 2 q ¯ 2 + 1 3 q ¯ ( p ρ ) 2 p C 1 ( 1 + ( 1 + 2 p 1 ) ( sup τ s 0 E ξ ( s ) p + C ) ) q ¯ ρ 2 p H 2 q ¯ 2 T C 2 p ρ q ¯ 2 p .

Substituting (81) into (79), we obtain

(82) J 21 C 1 E 0 t θ n e ¯ Δ ( s ) q ¯ d s + C ˜ 1 ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 .

Analogously to (79), we can show that

(83) J 22 C 1 E 0 t θ n e ¯ Δ ( s ) q ¯ d s + J 24 ,

where

J 24 = C 1 E 0 t θ n ( f Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) f Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) q ¯ + g Δ ( x Δ ( s ) , z ¯ Δ ( s ) ) g Δ ( x ¯ Δ ( s ) , y ¯ Δ ( s ) ) q ¯ ) d s .

Applying Assumption 8 , the Hölder inequality, and (48), we observe that

J 24 2 C 1 E 0 T ( H 2 ( 1 + x Δ ( s ) ρ + x ¯ Δ ( s ) ρ + z ¯ Δ ( s ) ρ + y ¯ Δ ( s ) ρ ) ( x Δ ( s ) x ¯ Δ ( s ) 2 + z ¯ Δ ( s ) y ¯ Δ ( s ) 2 ) ) q ¯ 2 d s 2 C 1 H 2 q ¯ 2 0 T E 5 q ¯ 2 1 1 + x Δ ( s ) q ¯ ρ 2 + x ¯ Δ ( s ) q ¯ ρ 2 + z ¯ Δ ( s ) q ¯ ρ 2 + y ¯ Δ ( s ) q ¯ ρ 2 2 q ¯ 2 1 ( x Δ ( s ) x ¯ Δ ( s ) q ¯ + z ¯ Δ ( s ) y ¯ Δ ( s ) q ¯ ) d s 2 q ¯ 2 5 q ¯ 2 1 C 1 H 2 q ¯ 2 0 T E 1 + x Δ ( s ) q ¯ ρ 2 + x ¯ Δ ( s ) q ¯ ρ 2 + z ¯ Δ ( s ) q ¯ ρ 2 + y ¯ Δ ( s ) q ¯ ρ 2 2 p q ¯ ρ q ¯ ρ 2 p × E ( x Δ ( s ) x ¯ Δ ( s ) q ¯ + z ¯ Δ ( s ) y ¯ Δ ( s ) q ¯ ) 2 p 2 p q ¯ ρ 2 p q ¯ ρ 2 p d s 2 q ¯ 2 + 1 2 p q ¯ ρ 2 p 5 q ¯ ( p ρ ) 2 p C 1 H 2 q ¯ 2 0 T [ 1 + E x Δ ( s ) p + E x ¯ Δ ( s ) p + E z ¯ Δ ( s ) p + E y ¯ Δ ( s ) p ] q ¯ ρ 2 p × E x Δ ( s ) x ¯ Δ ( s ) 2 p q ¯ 2 p q ¯ ρ + E z ¯ Δ ( s ) y ¯ Δ ( s ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p d s .

Now, (21), (22), and (23), together with Lemma 4, give

J 24 2 q ¯ 2 + 1 2 p q ¯ ρ 2 p 5 q ¯ ( p ρ ) 2 p C 1 H 2 q ¯ 2 0 T [ 1 + ( 3 + 2 p 1 ) sup τ r s E x Δ ( r ) p ] q ¯ ρ 2 p × E x Δ ( s ) x ¯ Δ ( s ) 2 p q ¯ 2 p q ¯ ρ + E z ¯ Δ ( s ) y ¯ Δ ( s ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p d s C 24 0 T E x Δ ( s ) x ¯ Δ ( s ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p + E z ¯ Δ ( s ) y ¯ Δ ( s ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p d s ,

where

C 24 = 2 q ¯ 2 5 q ¯ ( p ρ ) 2 p C 1 H 2 q ¯ 2 [ 1 + ( 3 + 2 p 1 ) ( sup τ s 0 E ξ ( s ) p + C ) ] .

By Corollary 1, we have that for l > 1 ,

(84) J 24 C 24 T c Δ p q ¯ 2 p q ¯ ρ + c l Δ 2 p q ¯ 2 p q ¯ ρ l 1 2 l ( h ( Δ ) ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p + c ¯ Δ p q ¯ 2 p q ¯ ρ + c ¯ l Δ 2 p q ¯ 2 p q ¯ ρ l 1 2 l ( h ( Δ ) ) 2 p q ¯ 2 p q ¯ ρ 2 p q ¯ ρ 2 p m ¯ Δ q ¯ 2 + m ¯ l Δ q ¯ 2 2 p q ¯ l 4 p l ( h ( Δ ) ) q ¯ ,

where m ¯ = 2 2 p q ¯ ρ 2 p C 24 T ( c c ¯ ) 2 p q ¯ ρ 2 p and m ¯ l = 2 2 p q ¯ ρ 2 p C 24 T ( c l c ¯ l ) 2 p q ¯ ρ 2 p . Substituting (84) into (83), we obtain

(85) J 22 C 1 E 0 t θ n e ¯ Δ ( s ) q ¯ d s + m ¯ Δ q ¯ 2 + m ¯ l Δ q ¯ 2 2 p q ¯ l 4 p l ( h ( Δ ) ) q ¯ .

Then, substitution of the estimates (82) and (85) into (78) yields

(86) J 2 2 C 1 E 0 t θ n e ¯ Δ ( s ) q ¯ d s + C ˜ 1 ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + m ¯ Δ q ¯ 2 + m ¯ l Δ q ¯ 2 2 p q ¯ l 4 p l ( h ( Δ ) ) q ¯ .

Consequently, (77) and (86), together with (70), give

(87) E e ¯ Δ ( t θ n ) q ¯ 2 H 1 ( q ¯ 2 ) E 0 t θ n e ¯ Δ ( s ) q ¯ d s + 2 q ¯ 1 + 1 1 δ ¯ H 1 E 0 t θ n e Δ ( s ) q ¯ d s + 2 2 q ¯ 1 H 1 T s ˇ Δ q ¯ 2 + s ˇ ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + s ˇ l Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ + 2 C 1 E 0 t θ n e ¯ Δ ( s ) q ¯ d s + C ˜ 1 ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + m ¯ Δ q ¯ 2 + m ¯ l Δ q ¯ 2 2 p q ¯ l 4 p l ( h ( Δ ) ) q ¯ 2 ( H 1 ( q ¯ 2 ) + C 1 ) E 0 t θ n e ¯ Δ ( s ) q ¯ d s + 2 q ¯ 1 + 1 1 δ ¯ H 1 E 0 t θ n e Δ ( s ) q ¯ d s + ( 2 2 q ¯ 1 s ˇ H 1 T + C ˜ 1 ) ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + ( 2 2 q ¯ 1 s ˇ H 1 T + m ¯ ) Δ q ¯ 2 + ( 2 2 q ¯ 1 s ˇ l H 1 T + m ¯ l ) Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ .

Based on Assumption 2 , Corollary 2, and (76), expression (87) can be estimated as

E e ¯ Δ ( t θ n ) q ¯ 2 q ¯ H 1 ( q ¯ 2 ) + C 1 + 1 + 1 1 δ ¯ H 1 E 0 t θ n e Δ ( s ) q ¯ d s + 2 q ¯ ( H 1 ( q ¯ 2 ) + C 1 ) E 0 t θ n u ( x ( s δ ( s ) ) ) u Δ ( z ¯ ( s ) ) q ¯ d s + ( 2 2 q ¯ 1 s ˇ H 1 T + C ˜ 1 ) ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + ( 2 2 q ¯ 1 s ˇ H 1 T + m ¯ ) Δ q ¯ 2 + ( 2 2 q ¯ 1 s ˇ l H 1 T + m ¯ l ) Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ 2 q ¯ H 1 ( q ¯ 2 ) + C 1 + 1 + 1 1 δ ¯ H 1 E 0 t θ n e Δ ( s ) q ¯ d s + 2 q ¯ ( H 1 ( q ¯ 2 ) + C 1 ) k q ¯ 0 T E x ( s δ ( s ) ) π Δ ( z ¯ ( s ) ) q ¯ d s + ( 2 2 q ¯ 1 s ˇ H 1 T + C ˜ 1 ) ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + ( 2 2 q ¯ 1 s ˇ H 1 T + m ¯ ) Δ q ¯ 2 + ( 2 2 q ¯ 1 s ˇ l H 1 T + m ¯ l ) Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ 2 q ¯ H 1 ( q ¯ 2 ) + C 1 + 1 + 1 1 δ ¯ H 1 E 0 t θ n e Δ ( s ) q ¯ d s + ( 2 2 q ¯ 1 s ˇ H 1 T + C ˜ 1 + 2 q ¯ ( H 1 ( q ¯ 2 ) + C 1 ) k q ¯ s ˇ T ) ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + ( 2 2 q ¯ 1 s ˇ H 1 T + m ¯ + 2 q ¯ ( H 1 ( q ¯ 2 ) + C 1 ) k q ¯ s ˇ T ) Δ q ¯ 2 + ( 2 2 q ¯ 1 s ˇ l H 1 T + m ¯ l + 2 q ¯ ( H 1 ( q ¯ 2 ) + C 1 ) k q ¯ s ˇ l T ) Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ .

Hence, we obtain that for l > 1 ,

(88) E e ¯ Δ ( t θ n ) q ¯ G 0 t E e Δ ( u θ n ) q ¯ d s + G ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + G Δ q ¯ 2 + G l Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ ,

where

G = max 2 q ¯ H 1 ( q ¯ 2 ) + C 1 + 1 + 1 1 δ ¯ H 1 , 2 2 q ¯ 1 s ˇ H 1 T + C ˜ 1 + 2 q ¯ ( H 1 ( q ¯ 2 ) + C 1 ) k q ¯ s ˇ T , 2 2 q ¯ 1 s ˇ H 1 T + m ¯ + 2 q ¯ ( H 1 ( q ¯ 2 ) + C 1 ) k q ¯ s ˇ T , G l = 2 2 q ¯ 1 s ˇ l H 1 T + m ¯ l + 2 q ¯ ( H 1 ( q ¯ 2 ) + C 1 ) k q ¯ s ˇ l T .

In the sequel, we will use the elementary inequality

(89) a + b p ( 1 + ε 1 ) p 1 a p + 1 + ε 1 ε 1 p 1 b p , a , b R , p > 1 .

Based on the fact that the solutions x and x Δ satisfy the same initial condition, from Assumption 2 , for any ε 1 , ε 2 > 0 , we obtain

sup 0 s t E e Δ ( s θ n ) q ¯ ( 1 + ε 1 ) q ¯ 1 sup 0 s t E e ¯ Δ ( s θ n ) q ¯ + 1 + ε 1 ε 1 q ¯ 1 k q ¯ sup 0 s t E x ( s θ n δ ( s θ n ) ) π Δ ( z ¯ Δ ( s θ n ) ) q ¯ ( 1 + ε 1 ) q ¯ 1 sup 0 s t E e ¯ Δ ( s θ n ) q ¯ + 1 + ε 1 ε 1 q ¯ 1 k q ¯ ( ( 1 + ε 2 ) q ¯ 1 sup 0 s t E x ( s θ n δ ( s θ n ) ) x Δ ( s θ n δ ( s θ n ) ) q ¯ + 1 + ε 2 ε 2 q ¯ 1 sup 0 s t E x Δ ( s θ n δ ( s θ n ) ) π Δ ( z ¯ Δ ( s θ n ) ) q ¯ ( 1 + ε 1 ) q ¯ 1 sup 0 s t E e ¯ Δ ( s θ n ) q ¯ + 1 + ε 1 ε 1 q ¯ 1 k q ¯ ( 1 + ε 2 ) q ¯ 1 sup 0 s t E e Δ ( s θ n ) q ¯ + 1 + ε 2 ε 2 q ¯ 1 sup 0 s t E x Δ ( s δ ( s ) ) π Δ ( z ¯ Δ ( s ) ) q ¯ .

So, for ε 1 = 1 1 k > k 1 k , ε 2 = ε 1 k ( 1 + ε 1 ) 1 = 1 k ( 2 k ) 1 and based on Corollary 2 and (76), we observe that for l > 1 ,

sup 0 s t E e Δ ( s θ n ) q ¯ ( 2 k ) q ¯ 1 ( 1 k ) q ¯ sup 0 s t E e ¯ Δ ( s θ n ) q ¯ + k q ¯ ( 2 k ) q ¯ 1 ( 1 k ) 2 q ¯ 1 s ˇ Δ q ¯ 2 + s ˇ ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + s ˇ l Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ .

Substituting the previous estimate into (88), we obtain

(90) E e ¯ Δ ( t θ n ) q ¯ G ( 2 k ) q ¯ 1 ( 1 k ) q ¯ 0 t sup 0 s t E e ¯ Δ ( s θ n ) q ¯ d s + G s ˇ T k q ¯ ( 2 k ) q ¯ 1 ( 1 k ) 2 q ¯ 1 + 1 Δ q ¯ 2 + G s ˇ T k q ¯ ( 2 k ) q ¯ 1 ( 1 k ) 2 q ¯ 1 + 1 ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + s ˇ l G T k q ¯ ( 2 k ) q ¯ 1 ( 1 k ) 2 q ¯ 1 + G l Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ .

Now, the application of the Gronwall-Bellman lemma to (90) yields

(91) sup 0 t T E e Δ ( t θ n ) q ¯ D ( μ 1 ( h ( Δ ) ) ) 2 p ( 2 + ρ ) q ¯ 2 + D Δ q ¯ 2 + D l Δ q ¯ l 1 2 l ( h ( Δ ) ) q ¯ ,

where

(92) D = G s ˇ T k q ¯ ( 2 k ) q ¯ 1 ( 1 k ) 2 q ¯ 1 + 1 e ( 2 k ) q ¯ 1 ( 1 k ) q ¯ G T , D l = s ˇ l G T k q ¯ ( 2 k ) q ¯ 1 ( 1 k ) 2 q ¯ 1 + G l e ( 2 k ) q ¯ 1 ( 1 k ) q ¯ G T .

Using the well-known Fatou lemma, we can let n to obtain the desired assertion (64). The other assertion (65) follows from (64) and Corollary 7, for constants

(93) D = 2 q ¯ 1 D ( D + 2 q ¯ 1 c ) , and D l = D l + 2 q ¯ 1 c l .

Finally, if μ is defined by (66) then its inverse function is given by μ 1 ( u ) = ( u H 3 ) 2 ρ + 2 . So, based on the definition of function h ( Δ ) = h ˆ Δ ε and from (64), we obtain that for l > 1 ,

(94) sup 0 t T E e Δ ( t ) q ¯ D ˜ Δ ε 2 p ρ + 2 q ¯ + Δ q ¯ 2 + Δ q ¯ 2 1 2 l ε q ¯ ,

where

(95) D ˜ = D H 3 h ˆ 2 p 2 + ρ q ¯ D D l h ˆ q ¯ .

Therefore, (94) is the required assertion (67). Similarly, we can show (68). The proof is therefore complete.□

The following theorem shows that the rate of L q ¯ -convergence could be close arbitrarily to q ¯ 2 .

Theorem 3

Let the conditions of Theorem 2 hold. Then, for any q ¯ [ 2 , q ) and any ε 0 , 1 ε 0 4 ,

(96) sup 0 t T E x ( t ) x Δ ( t ) q ¯ O Δ q ¯ ( 1 4 ε ) 2 ,

(97) sup 0 t T E x ( t ) x ¯ Δ ( t ) q ¯ O Δ q ¯ ( 1 4 ε ) 2 .

Proof

Choosing p sufficiently large for

(98) ε ( 2 p ( 2 + ρ ) q ¯ ) 2 + ρ > q ¯ ( 1 4 ε ) 2 ,

and l > 1 2 ε q ¯ , we can obtain the assertions from (96) and (97) easily.□

Let us discuss an example to illustrate our theory.

Example 1

Consider the following one-dimensional neutral stochastic differential equation with time-dependent delay:

(99) d x ( t ) + 1 27 sin ( x ( t δ ( t ) ) ) = f ( x ( t ) , x ( t δ ( t ) ) ) d t + g ( x ( t ) , x ( t δ ( t ) ) ) d B ( t ) , t [ 0 , T ] ,

where T = 1 , with the initial condition ξ ( θ ) = 1 , θ [ τ , 0 ] , τ = 2 . The delay function is defined as δ ( t ) = 1 1 4 sin t , t [ 0 , 1 ] and

f ( x , y ) = x + 1 27 sin y x + 1 27 sin y 3 , g ( x , y ) = x + y 1 + y 2 , x , y R .

Clearly, the coefficients f and g are locally Lipschitz continuous, namely, they satisfy Assumption 1 , while

u ( x ) = 1 27 sin x satisfies Assumption 2 for k = 1 27 . Since δ ( t ) 1 4 = δ ¯ , Assumption 3 holds. Moreover, for any p 2 and any a ( 0 , 1 ] , we have

( x a u ( y ) ) T f ( x , y ) + p 1 2 g ( x , y ) 2 x + a 3 3 sin y x + 1 3 3 sin y x 3 + 1 3 2 x 2 sin y + 1 3 5 x sin 2 y + 1 3 9 sin 3 y + ( p 1 ) x 2 + y 2 ( 1 + y 2 ) 2 55 + a 54 x 2 + 29 a + 27 1,458 y 2 x 4 + a + 3 3 3 x 3 + 3 a + 1 3 9 x + ( p 1 ) ( x 2 + y 2 ) K ( 1 + x 2 + a y 2 ) ,

where

K = 55 + a 54 + p 1 29 1,458 + 27 1,458 a + p 1 a a ˜ , a ˜ = sup u 0 u 4 + a + 3 3 3 u 3 + 3 a + 1 3 9 u .

Thus, Assumption 4 holds for any p 2 and any a ( 0 , 1 ] . Particularly, for a = 1 , we find that Assumption 4 is fulfilled as well. The initial condition ξ ( θ ) = 1 , θ [ τ , 0 ] satisfies Assumption 5 . Noting that

δ ( t ) δ ( s ) 1 4 t s , t , s 0 ,

we find that 6 holds with η = 1 4 . Also, since k ( 3 + η ) = 1 9 < 1 , condition (35) holds.

Furthermore, it is easy to show that for any q > 2 ,

( x u ( y ) x ¯ + u ( y ¯ ) ) T ( f ( x , y ) f ( x ¯ , y ¯ ) ) + q 1 2 g ( x , y ) g ( x ¯ , y ¯ ) 2 x x ¯ + 1 27 ( sin y sin y ¯ ) 2 1 x + 1 27 sin y 2 x + 1 27 sin y x ¯ + 1 27 sin y ¯ x ¯ + 1 27 sin y ¯ 2 + ( q 1 ) x x ¯ 2 + ( q 1 ) y 1 + y 2 y ¯ 1 + y ¯ 2 2 2 x x ¯ 2 + 2 2 7 2 y y ¯ 2 + ( q 1 ) x x ¯ 2 + ( q 1 ) y y ¯ 2 ( q + 1 ) ( x x ¯ 2 + y y ¯ 2 ) .

Thus, Assumption 7 holds for any q > 2 . Moreover, we have

f ( x , y ) f ( x ¯ , y ¯ ) 2 2 x x ¯ 2 + 1 2 7 2 y y ¯ 2 1 1 2 x + 1 27 sin y 2 1 2 x ¯ + 1 27 sin y ¯ 2 2 2 ( x x ¯ 2 + y y ¯ 2 ) ( 1 + x ρ + y ρ + x ¯ ρ + y ¯ ρ ) .

Thus, Assumption 8 is satisfied, for example, with ρ = 1 . Obviously, we can choose values of p and q , such that relation 2 p > ( 2 + ρ ) q holds. To apply Theorem 2, we still need to design functions μ and h . Note that from (49), we obtain, for all r 1 .

sup x y r ( f ( x , y ) g ( x , y ) u ( y ) ) 4 r 3 2 .

So, we can have μ ( r ) = 4 r 3 2 and its inverse function μ 1 ( r ) = ( r 4 ) 2 3 for r 4 . For ε 0 , 1 ε 0 4 , we define h ( Δ ) = Δ ε for Δ > 0 . Now, for any q ¯ 2 , we can choose p sufficiently large for

ε ( 2 p ( 2 + ρ ) q ¯ ) 2 + ρ > q ¯ ( 1 4 ε ) 2 .

Therefore, it follows from Theorem 3 that the truncated EM solutions of equation (99) satisfy relations (96) and (97), i.e., the order of L q ¯ -convergence can be arbitrarily close to q ¯ 2 .

Now, we can choose ε 0 = 1 2 and ε = 1 ε 0 8 = 1 16 . It is easy to see that condition (10) becomes

Δ * 2 32 2.3283 × 1 0 10 .

For such a small step size Δ ( 0 , Δ * ] , the results from [24] hold. In this article, the step size Δ can be any number in ( 0 , 1 ) . The advantage of our results is even more clear if we choose ε = 1 ε 0 64 = 1 128 . In this case, condition (10) becomes

Δ * 2 256 8.6362 × 1 0 78 .

This is almost impossible so results from [24] are much more restrictive. However, results in this article can still be applied. Also, we see that for ε = 1 128 , relations (96) and (97) yield

sup 0 t T E x ( t ) x Δ ( t ) q ¯ O Δ q ¯ 2 q ¯ 64 , sup 0 t T E x ( t ) x ¯ Δ ( t ) q ¯ O Δ q ¯ 2 q ¯ 64 .

So, Theorem 3 is not only applicable in this situation but also shows that the truncated EM solution x Δ ( t ) defined using μ ( r ) = 4 r 3 2 and h ( Δ ) = Δ 1 128 has a better strong convergence rate to the true solution of equation (99) than that using μ ( r ) = 4 r 3 2 and h ( Δ ) = Δ 1 4 .

We choose Δ = 0.0078 , ε 0 = 1 2 and ε = 1 ε 0 64 = 1 128 to simulate several trajectories of the truncated EM solution of equation (99), which can be seen in Figure 1.

On the other hand, on the basis of Theorem 3, for q ¯ = 2 and ε = 1 10 , we have that

E x ( T ) x ¯ Δ ( T ) q ¯ D ˜ Δ q ¯ 1 2 2 ε ,

where D ˜ is given by (95). In what follows, the main aim is to compare the rate of convergence 1 2 2 ε from this article with 1 2 , i.e., to compare the approximation errors D ˜ Δ q ¯ 1 2 2 ε and D ˜ Δ q ¯ 2 . So, taking logarithms, we obtain that

ln E x ( T ) x ¯ Δ ( T ) ln D ˜ q ¯ + 1 2 2 ε ln Δ .

Therefore, the problem reduces to comparing the lines a 1 2 , Δ and a 1 2 2 ε , Δ , where a ( x , Δ ) = x ln Δ + ln D ˜ q ¯ , for a fixed value of Δ ( 0 , 1 ) and then, making calculations about the approximation errors for small enough Δ . In order to do this, we plotted the lines a 1 2 , Δ and a 1 2 2 ε , Δ for Δ ( 0 , 1 ) in Figure 2, where z = ln Δ , q ¯ = 2 , ε = 0.1 . Thus, when Δ is fixed and treated like a parameter, the increment of that function can be represented as

a 1 2 2 ε , Δ a 1 2 , Δ = ln Δ ( 2 ε ) ,

which tends to + as Δ 0 . This result corresponds to the graphical representation of lines a 1 2 , Δ and a 1 2 2 ε , Δ in Figure 2.

However, in order to make a conclusion to the initial problem, the function b ( x , Δ ) = e a ( x , Δ ) is of interest. So, on the basis of the mean value theorem, treating Δ like a parameter, we have that, for some θ ( 0 , 1 ) , the increment of that function can be represented as

b 1 2 2 ε , Δ b 1 2 , Δ = D ˜ 1 q ¯ Δ 1 2 2 ( 1 θ ) ε ln Δ ( 2 ε ) ,

which tends to 0 when Δ 0 . Therefore, we can conclude that, for sufficiently small Δ , the approximation error O Δ q ¯ 1 2 2 ε is getting closer to O ( Δ q ¯ 2 ) . It should be emphasized that when calculating the constant D ˜ , constants c ˜ , c ˜ l , c ¯ , c ¯ l , c , c l , C l , N ¯ , M l , Q l , S l , and C were used from article [24], but taking into account Lemmas 5 and 6, such that 3 k from those constants is substituted by k , i.e.,

c ˜ = C ξ 1 γ , C ξ = 0 , γ = 1 + ε ¯ 1 p ˆ 1 p ˆ 1 k p ˆ ε ¯ ( 3 + η ) p ˆ , η = 1 4 , ε ¯ = 1 , p ˆ = 3 , k = 1 27 , c ˜ l = 2 p ˆ 1 1 γ 1 + ε ¯ 1 p ˆ 1 p ˆ 1 1 + m p ˆ 2 ( T ( 2 p ˆ l 1 ) ! ! ) 1 2 l , m = 1 , l = 6 , T = 1 , c ¯ = ( 3 + η ) p ˆ c ˜ , c ¯ l = ( 3 + η ) p ˆ c ˜ l , c = 3 p 1 ˆ k p ˆ ( 3 + η ) p ˆ c ˜ , c l = 3 p 1 ˆ k p ˆ ( 3 + η ) p ˆ c ˜ l + 1 + p ˆ ( p ˆ 1 ) 2 p ˆ 2 ,

C l = 2 p 1 2 T ( c + k p c ¯ ) 1 2 + ( c l + k p c ¯ l ) 1 2 , p = 13 , N ¯ = 8 K ˆ [ 4 K ˆ ( p 2 ) T ] p 2 p 3 p 2 1 ( 1 k ) p , K ˆ = 3 2 K 1 1 μ 1 ( h ( Δ ) ) , M l = 4 K ˆ [ 4 K ˆ ( p 2 ) T ] p 2 p 3 p 2 1 T + 4 ( 4 ( p 2 ) T ) p 2 2 C l ( 1 k ) p , Q l = k 1 k + 2 p ( 1 + k p ) ( 1 k ) p sup τ s 0 E ξ ( s ) p + M l = k 1 k + 2 p ( 1 + k p ) ( 1 k ) p + M l , S l = Q l e N T ¯ and C = S l S l p 2 .

Also, for the computational convenience, the constant D l from (93) was majorized by a larger constant 2 D l , where D l is given by (92).

Moreover, for fixed parameters p = 13 , q ¯ = 2 , ε 0 = 0.001 , and different values of ε , we can conclude that the slope of the line a 1 2 2 ε , Δ is closer to the slope of a 1 2 , Δ for smaller values of ε (Figure 3).

Additionally, if we increase the values of p and q ¯ , ( p = 20 and q ¯ = 3 ), then for different values of ε , we obtain the same conclusion. As ε decreases, the lines a 1 2 , Δ and a 1 2 2 ε , Δ become closer to each other (Figure 4).

Figure 1 
               Trajectories of the truncated EM solution with 
                     
                        
                        
                           Δ
                           =
                           0.0078
                        
                        \Delta =0.0078
                     
                  .
Figure 1

Trajectories of the truncated EM solution with Δ = 0.0078 .

Figure 2 
               Lines 
                     
                        
                        
                           a
                                                         
                              
                                 
                                    
                                       1
                                    
                                    
                                       2
                                    
                                 
                                 ,
                                 Δ
                              
                           
                        
                        a\left(\phantom{\rule[-0.75em]{}{0ex}},\frac{1}{2},\Delta \right)
                     
                   (orange line) and 
                     
                        
                        
                           a
                                                         
                              
                                 
                                    
                                       1
                                    
                                    
                                       2
                                    
                                 
                                 −
                                 2
                                 ε
                                 ,
                                 Δ
                              
                           
                        
                        a\left(\phantom{\rule[-0.75em]{}{0ex}},\frac{1}{2}-2\varepsilon ,\Delta \right)
                     
                   (blue line) for 
                     
                        
                        
                           Δ
                           ∈
                           
                              (
                              
                                 0
                                 ,
                                 1
                              
                              )
                           
                           ,
                           
                              
                                 q
                              
                              
                                 ¯
                              
                           
                           =
                           2
                           ,
                           ε
                           =
                           0.1
                        
                        \Delta \in \left(0,1),\bar{q}=2,\varepsilon =0.1
                     
                  .
Figure 2

Lines a 1 2 , Δ (orange line) and a 1 2 2 ε , Δ (blue line) for Δ ( 0 , 1 ) , q ¯ = 2 , ε = 0.1 .

Figure 3 
               Lines 
                     
                        
                        
                           a
                                                         
                              
                                 
                                    
                                       1
                                    
                                    
                                       2
                                    
                                 
                                 ,
                                 Δ
                              
                           
                        
                        a\left(\phantom{\rule[-0.75em]{}{0ex}},\frac{1}{2},\Delta \right)
                     
                   (orange line) and 
                     
                        
                        
                           a
                                                         
                              
                                 
                                    
                                       1
                                    
                                    
                                       2
                                    
                                 
                                 −
                                 2
                                 ε
                                 ,
                                 Δ
                              
                           
                        
                        a\left(\phantom{\rule[-0.75em]{}{0ex}},\frac{1}{2}-2\varepsilon ,\Delta \right)
                     
                   (blue line) for 
                     
                        
                        
                           Δ
                           ∈
                           
                              (
                              
                                 0
                                 ,
                                 1
                              
                              )
                           
                           ,
                           p
                           =
                           13
                           ,
                           
                              
                                 q
                              
                              
                                 ¯
                              
                           
                           =
                           2
                           ,
                           
                              
                                 ε
                              
                              
                                 0
                              
                           
                           =
                           0.001
                           ,
                        
                        \Delta \in \left(0,1),p=13,\bar{q}=2,{\varepsilon }_{0}=0.001,
                     
                   (a) 
                     
                        
                        
                           ε
                           =
                           0.2
                           ,
                        
                        \varepsilon =0.2,
                     
                   (b) 
                     
                        
                        
                           ε
                           =
                           0.17
                           ,
                        
                        \varepsilon =0.17,
                     
                   (c) 
                     
                        
                        
                           ε
                           =
                           0.14
                           ,
                        
                        \varepsilon =0.14,
                     
                   and (d) 
                     
                        
                        
                           ε
                           =
                           0.1
                           .
                        
                        \varepsilon =0.1.
Figure 3

Lines a 1 2 , Δ (orange line) and a 1 2 2 ε , Δ (blue line) for Δ ( 0 , 1 ) , p = 13 , q ¯ = 2 , ε 0 = 0.001 , (a) ε = 0.2 , (b) ε = 0.17 , (c) ε = 0.14 , and (d) ε = 0.1 .

Figure 4 
               Lines 
                     
                        
                        
                           a
                                                         
                              
                                 
                                    
                                       1
                                    
                                    
                                       2
                                    
                                 
                                 ,
                                 Δ
                              
                           
                        
                        a\left(\phantom{\rule[-0.75em]{}{0ex}},\frac{1}{2},\Delta \right)
                     
                   (orange line) and 
                     
                        
                        
                           a
                                                         
                              
                                 
                                    
                                       1
                                    
                                    
                                       2
                                    
                                 
                                 −
                                 2
                                 ε
                                 ,
                                 Δ
                              
                           
                        
                        a\left(\phantom{\rule[-0.75em]{}{0ex}},\frac{1}{2}-2\varepsilon ,\Delta \right)
                     
                   (blue line) for 
                     
                        
                        
                           Δ
                           ∈
                           
                              (
                              
                                 0
                                 ,
                                 1
                              
                              )
                           
                           ,
                           p
                           =
                           20
                           ,
                           
                              
                                 q
                              
                              
                                 ¯
                              
                           
                           =
                           3
                           ,
                           
                              
                                 ε
                              
                              
                                 0
                              
                           
                           =
                           0.001
                           ,
                        
                        \Delta \in \left(0,1),p=20,\bar{q}=3,{\varepsilon }_{0}=0.001,
                     
                   (a) 
                     
                        
                        
                           ε
                           =
                           0.2
                           ,
                        
                        \varepsilon =0.2,
                     
                   (b) 
                     
                        
                        
                           ε
                           =
                           0.17
                           ,
                        
                        \varepsilon =0.17,
                     
                   (c) 
                     
                        
                        
                           ε
                           =
                           0.14
                           ,
                        
                        \varepsilon =0.14,
                     
                   and (d) 
                     
                        
                        
                           ε
                           =
                           0.1
                        
                        \varepsilon =0.1
                     
                  .
Figure 4

Lines a 1 2 , Δ (orange line) and a 1 2 2 ε , Δ (blue line) for Δ ( 0 , 1 ) , p = 20 , q ¯ = 3 , ε 0 = 0.001 , (a) ε = 0.2 , (b) ε = 0.17 , (c) ε = 0.14 , and (d) ε = 0.1 .

4 Conclusions

On the basis of the previously presented results, one can conclude that the rate of the L q convergence of the truncated EM method is arbitrarily close to 1 2 , i.e., to the rate of the L q convergence of the classical EM method. This article represents the contribution to the analysis of stochastic differential equations with highly nonlinear coefficients (as illustrated by the example), which arise in modeling many real-life phenomena. As one can observe, one of the advantages of the method considered in this article is the fact that it is explicit, which is important from both theoretical and practical aspects. Namely, the theoretical approach does not require additional conditions that guarantee the existence and uniqueness of numerical solutions, while numerical simulations are less demanding and go faster in comparison with the implicit numerical methods.

On the other hand, when one applies this method, one should pay attention for which value of p there exist moments of both exact and approximate solutions, since the L q convergence theorem is proved using corresponding moment bounds and q is less than p . Also, from (98), one can observe that ε cannot be chosen arbitrarily small, i.e., it must

ε > q ¯ ( 2 + ρ ) 4 p + 2 q ¯ ( 2 + ρ ) ,

which affects the convergence rate of the method.

Possible directions of the future research could be the extensions of the presented results to neutral stochastic differential equations with time-distributed delay, with state-dependent delay, for example, with or without some additional noise, such as Poison jump, Lévy noise, or Markovian switching. Also, the future study could be based on determining some other sufficient conditions of the L q convergence of the method, i.e., on revealing some other class neutral stochastic differential equations for which the L q convergence of the method holds.

Acknowledgment

The author is grateful for the reviewer’s valuable comments that improved the manuscript.

  1. Funding information: This work was supported by Ministry of Education, Science and Technological Development of the Republic of Serbia, Contract Number: 451-03-66/2024-03/200124.

  2. Author contributions: The author confirms the sole responsibility for the conception of the study, presented results, and prepared manuscript.

  3. Conflict of interest: The author states no conflict of interest.

  4. Ethical approval: The conducted research is not related to either human or animal use.

  5. Data availability statement: Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

References

[1] M. Wu, W. Cao, and F. Wu, An averaging principle for fast-slow-coupled neutral stochastic differential equations with time-varying delay, Appl. Math. Optim. 88 (2023), no. 82, DOI: https://doi.org/10.1007/s00245-023-10057-7. 10.1007/s00245-023-10057-7Search in Google Scholar

[2] N. Jacob, Y. Wang, and C. Yuan, Stochastic differential delay equations with jumps, under nonlinear growth condition, Stochastics 81 (2009), no. 6, 571–588, DOI: https://doi.org/10.1080/17442500903251832. 10.1080/17442500903251832Search in Google Scholar

[3] Q. Luo, X. Mao, and Y. Shen, New criteria on exponential stability of neutral stochastic differential delay equations, Syst. Control Lett. 55 (2006), no. 10, 826–834, DOI: https://doi.org/10.1016/j.sysconle.2006.04.005. 10.1016/j.sysconle.2006.04.005Search in Google Scholar

[4] X. Mao, Asymptotic properties of neutral stochastic differential delay equations, Stochastics Stochastics Reports 68 (2000), no. 3–4, 273–295, DOI: https://doi.org/10.1080/17442500008834226. 10.1080/17442500008834226Search in Google Scholar

[5] X. Mao and M. J. Rassias, Khasminskii-type theorems for stochastic differential delay equations, Stoch. Anal. Appl. 23 (2005), no. 5, 1045–1069, DOI: https://doi.org/10.1080/07362990500118637. 10.1080/07362990500118637Search in Google Scholar

[6] Q. Guo, X. Mao, and R. Yue, The truncated Euler-Maruyama method for stochastic differential delay equations, Numer. Algorithms 78 (2018), 599–624, DOI: https://doi.org/10.1007/s11075-017-0391-0. 10.1007/s11075-017-0391-0Search in Google Scholar

[7] L. Hu, X. Li, and X. Mao, Convergence rate and stability of the truncated Euler-Maruyama method for stochastic differential equations, J. Comput. Appl. Math. 337 (2018), 274–289, DOI: https://doi.org/10.1016/j.cam.2018.01.017. 10.1016/j.cam.2018.01.017Search in Google Scholar

[8] P. E. Kloeden and E. Platen, Numerical Solutions of Stochastic Differential Equations, Springer, Berlin, 1992. 10.1007/978-3-662-12616-5Search in Google Scholar

[9] Q. Guo, W. Liu, and X. Mao, A note on the partially truncated Euler-Maruyama method, Appl. Numer. Math. 130 (2018), 157–170, DOI: https://doi.org/10.1016/j.apnum.2018.04.004. 10.1016/j.apnum.2018.04.004Search in Google Scholar

[10] X. Mao, The truncated Euler-Maruyama method for stochastic differential equations, J. Comput. Appl. Math. 290 (2015), 370–384, DOI: https://doi.org/10.1016/j.cam.2015.06.002. 10.1016/j.cam.2015.06.002Search in Google Scholar

[11] L. Tan and C. Yuan, Convergence rates of truncated EM scheme for NSDDEs (2018), arXiv: arXiv:1801.05952v1, DOI: https://doi.org/10.48550/arXiv.1801.05952. Search in Google Scholar

[12] F. Wu and X. Mao, Numerical solutions of neutral stochastic functional differential equations, SIAM J. Numer. Anal. 46 (2008), 1821–1841, DOI: https://doi.org/10.1137/07069702. 10.1137/070697021Search in Google Scholar

[13] Z. Bao, J. Tang, Y. Shen, and W. Liu, Equivalence of pth moment stability between stochastic differential delay equations and their numerical methods, Statist. Probab. Lett. 168 (2021), 108952, DOI: https://doi.org/10.1016/j.spl.2020.108952. 10.1016/j.spl.2020.108952Search in Google Scholar

[14] Z. Yu, Almost surely asymptotic stability of exact and numerical solutions for neutral stochastic pantograph equations, Abstr. Appl. Anal. 2011 (2011), 143079, DOI: https://doi.org/10.1155/2011/143079. 10.1155/2011/143079Search in Google Scholar

[15] X. Mao, Numerical solutions of stochastic differential delay equations under the generalized Khasminskii-type conditions, Appl. Math. Comput. 217 (2011), 5512–5524, DOI: https://doi.org/10.1016/j.amc.2010.12.023. 10.1016/j.amc.2010.12.023Search in Google Scholar

[16] X. Mao, Numerical solutions of stochastic differential delay equations under local Lipschitz condition, J. Comput. Appl. Math. 151 (2003), 215–227, DOI: https://doi.org/10.1016/S0377-0427(02)00750-1. 10.1016/S0377-0427(02)00750-1Search in Google Scholar

[17] M. Milošević, Highly nonlinear neutral stochastic differential equations with time-dependent delay and the Euler-Maruyama method, Math. Comput. Model. 54 (2011), 2235–2251, DOI: https://doi.org/10.1016/j.mcm.2011.05.033. 10.1016/j.mcm.2011.05.033Search in Google Scholar

[18] H. Liu, F. Wu, and M. Wu, The Tamed Euler-Maruyama approximation of Mckean-Vlasov stochastic differential equations and asymptotic error analysis, Discrete Contin. Dyn. Syst. Ser. S 16 (2023), no. 5, 1014–1040, DOI: https://doi.org/10.3934/dcdss.2023029. 10.3934/dcdss.2023029Search in Google Scholar

[19] H. Liu, B. Shi, and F. Wu, Tamed Euler-Maruyama approximation of McKean-Vlasov stochastic differential equations with super-linear drift and Holder diffusion coefficients, Appl. Numer. Math. 183 (2023), 56–85, DOI: https://doi.org/10.1016/j.apnum.2022.08.012. 10.1016/j.apnum.2022.08.012Search in Google Scholar

[20] Y. Li, X. Mao, Q. Song, F. Wu, and G. Yin, Strong convergence of Euler-Maruyama schemes for McKean-Vlasov stochastic differential equations under local Lipschitz conditions of state variables, IMA J. Numer. Anal. 43 (2023), no. 2, 1001–1035, DOI: https://doi.org/10.1093/imanum/drab107. 10.1093/imanum/drab107Search in Google Scholar

[21] H. Yang, F. Wu, P. E. Kloeden, and X. Mao, The truncated Euler-Maruyama method for stochastic differential equations with Holder diffusion coefficients, J. Comput. Appl. Math. 366 (2020), 112379, DOI: https://doi.org/10.1016/j.cam.2019.112379. 10.1016/j.cam.2019.112379Search in Google Scholar

[22] M. Hutzenthaler, A. Jentzen, and P. E. Kloeden, Strong and weak divergence in finite time of Euler’s method for stochastic differential equations with non-globally Lipschitz continuous coefficients, Proc. R. Soc. A 467 (2011), no. 2130, 1563–1576, DOI: https://doi.org/10.1098/rspa.2010.0348. 10.1098/rspa.2010.0348Search in Google Scholar

[23] W. Zhang, Convergence rate of the truncated Euler-Maruyama method of neutral stochastic differential delay equations with Markovian switching, J. Comput. Math. 38 (2020), 874–904, DOI: https://doi:10.4208/jcm.1906-m2018-0237. 10.4208/jcm.1906-m2018-0237Search in Google Scholar

[24] A. Petrović and M. Milošević, The truncated Euler-Maruyama method for highly nonlinear neutral stochastic differential equations with time-dependent delay, Filomat 35 (2021), no. 7, 2457–2484. 10.2298/FIL2107457PSearch in Google Scholar

Received: 2024-01-15
Revised: 2024-07-09
Accepted: 2024-07-15
Published Online: 2024-08-14

© 2024 the author(s), published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Articles in the same Issue

  1. Special Issue on Contemporary Developments in Graph Topological Indices
  2. On the maximum atom-bond sum-connectivity index of graphs
  3. Upper bounds for the global cyclicity index
  4. Zagreb connection indices on polyomino chains and random polyomino chains
  5. On the multiplicative sum Zagreb index of molecular graphs
  6. The minimum matching energy of unicyclic graphs with fixed number of vertices of degree two
  7. Special Issue on Convex Analysis and Applications - Part I
  8. Weighted Hermite-Hadamard-type inequalities without any symmetry condition on the weight function
  9. Scattering threshold for the focusing energy-critical generalized Hartree equation
  10. (pq)-Compactness in spaces of holomorphic mappings
  11. Characterizations of minimal elements of upper support with applications in minimizing DC functions
  12. Some new Hermite-Hadamard-type inequalities for strongly h-convex functions on co-ordinates
  13. Global existence and extinction for a fast diffusion p-Laplace equation with logarithmic nonlinearity and special medium void
  14. Extension of Fejér's inequality to the class of sub-biharmonic functions
  15. On sup- and inf-attaining functionals
  16. Regularization method and a posteriori error estimates for the two membranes problem
  17. Rapid Communication
  18. Note on quasivarieties generated by finite pointed abelian groups
  19. Review Articles
  20. Amitsur's theorem, semicentral idempotents, and additively idempotent semirings
  21. A comprehensive review of the recent numerical methods for solving FPDEs
  22. On an Oberbeck-Boussinesq model relating to the motion of a viscous fluid subject to heating
  23. Pullback and uniform exponential attractors for non-autonomous Oregonator systems
  24. Regular Articles
  25. On certain functional equation related to derivations
  26. The product of a quartic and a sextic number cannot be octic
  27. Combined system of additive functional equations in Banach algebras
  28. Enhanced Young-type inequalities utilizing Kantorovich approach for semidefinite matrices
  29. Local and global solvability for the Boussinesq system in Besov spaces
  30. Construction of 4 x 4 symmetric stochastic matrices with given spectra
  31. A conjecture of Mallows and Sloane with the universal denominator of Hilbert series
  32. The uniqueness of expression for generalized quadratic matrices
  33. On the generalized exponential sums and their fourth power mean
  34. Infinitely many solutions for Schrödinger equations with Hardy potential and Berestycki-Lions conditions
  35. Computing the determinant of a signed graph
  36. Two results on the value distribution of meromorphic functions
  37. Zariski topology on the secondary-like spectrum of a module
  38. On deferred f-statistical convergence for double sequences
  39. About j-Noetherian rings
  40. Strong convergence for weighted sums of (α, β)-mixing random variables and application to simple linear EV regression model
  41. On the distribution of powered numbers
  42. Almost periodic dynamics for a delayed differential neoclassical growth model with discontinuous control strategy
  43. A new distributionally robust reward-risk model for portfolio optimization
  44. Asymptotic behavior of solutions of a viscoelastic Shear beam model with no rotary inertia: General and optimal decay results
  45. Silting modules over a class of Morita rings
  46. Non-oscillation of linear differential equations with coefficients containing powers of natural logarithm
  47. Mutually unbiased bases via complex projective trigonometry
  48. Hyers-Ulam stability of a nonlinear partial integro-differential equation of order three
  49. On second-order linear Stieltjes differential equations with non-constant coefficients
  50. Complex dynamics of a nonlinear discrete predator-prey system with Allee effect
  51. The fibering method approach for a Schrödinger-Poisson system with p-Laplacian in bounded domains
  52. On discrete inequalities for some classes of sequences
  53. Boundary value problems for integro-differential and singular higher-order differential equations
  54. Existence and properties of soliton solution for the quasilinear Schrödinger system
  55. Hermite-Hadamard-type inequalities for generalized trigonometrically and hyperbolic ρ-convex functions in two dimension
  56. Endpoint boundedness of toroidal pseudo-differential operators
  57. Matrix stretching
  58. A singular perturbation result for a class of periodic-parabolic BVPs
  59. On Laguerre-Sobolev matrix orthogonal polynomials
  60. Pullback attractors for fractional lattice systems with delays in weighted space
  61. Singularities of spherical surface in R4
  62. Variational approach to Kirchhoff-type second-order impulsive differential systems
  63. Convergence rate of the truncated Euler-Maruyama method for highly nonlinear neutral stochastic differential equations with time-dependent delay
  64. On the energy decay of a coupled nonlinear suspension bridge problem with nonlinear feedback
  65. The limit theorems on extreme order statistics and partial sums of i.i.d. random variables
  66. Hardy-type inequalities for a class of iterated operators and their application to Morrey-type spaces
  67. Solving multi-point problem for Volterra-Fredholm integro-differential equations using Dzhumabaev parameterization method
  68. Finite groups with gcd(χ(1), χc (1)) a prime
  69. Small values and functional laws of the iterated logarithm for operator fractional Brownian motion
  70. The hull-kernel topology on prime ideals in ordered semigroups
  71. ℐ-sn-metrizable spaces and the images of semi-metric spaces
  72. Strong laws for weighted sums of widely orthant dependent random variables and applications
  73. An extension of Schweitzer's inequality to Riemann-Liouville fractional integral
  74. Construction of a class of half-discrete Hilbert-type inequalities in the whole plane
  75. Analysis of two-grid method for second-order hyperbolic equation by expanded mixed finite element methods
  76. Note on stability estimation of stochastic difference equations
  77. Trigonometric integrals evaluated in terms of Riemann zeta and Dirichlet beta functions
  78. Purity and hybridness of two tensors on a real hypersurface in complex projective space
  79. Classification of positive solutions for a weighted integral system on the half-space
  80. A quasi-reversibility method for solving nonhomogeneous sideways heat equation
  81. Higher-order nonlocal multipoint q-integral boundary value problems for fractional q-difference equations with dual hybrid terms
  82. Noetherian rings of composite generalized power series
  83. On generalized shifts of the Mellin transform of the Riemann zeta-function
  84. Further results on enumeration of perfect matchings of Cartesian product graphs
  85. A new extended Mulholland's inequality involving one partial sum
  86. Power vector inequalities for operator pairs in Hilbert spaces and their applications
  87. On the common zeros of quasi-modular forms for Γ+0(N) of level N = 1, 2, 3
  88. One special kind of Kloosterman sum and its fourth-power mean
  89. The stability of high ring homomorphisms and derivations on fuzzy Banach algebras
  90. Integral mean estimates of Turán-type inequalities for the polar derivative of a polynomial with restricted zeros
  91. Commutators of multilinear fractional maximal operators with Lipschitz functions on Morrey spaces
  92. Vector optimization problems with weakened convex and weakened affine constraints in linear topological spaces
  93. The curvature entropy inequalities of convex bodies
  94. Brouwer's conjecture for the sum of the k largest Laplacian eigenvalues of some graphs
  95. High-order finite-difference ghost-point methods for elliptic problems in domains with curved boundaries
  96. Riemannian invariants for warped product submanifolds in Q ε m × R and their applications
  97. Generalized quadratic Gauss sums and their 2mth power mean
  98. Euler-α equations in a three-dimensional bounded domain with Dirichlet boundary conditions
  99. Enochs conjecture for cotorsion pairs over recollements of exact categories
  100. Zeros distribution and interlacing property for certain polynomial sequences
  101. Random attractors of Kirchhoff-type reaction–diffusion equations without uniqueness driven by nonlinear colored noise
  102. Study on solutions of the systems of complex product-type PDEs with more general forms in ℂ2
  103. Dynamics in a predator-prey model with predation-driven Allee effect and memory effect
  104. A note on orthogonal decomposition of 𝔰𝔩n over commutative rings
  105. On the δ-chromatic numbers of the Cartesian products of graphs
  106. Binomial convolution sum of divisor functions associated with Dirichlet character modulo 8
  107. Commutator of fractional integral with Lipschitz functions related to Schrödinger operator on local generalized mixed Morrey spaces
  108. System of degenerate parabolic p-Laplacian
  109. Stochastic stability and instability of rumor model
  110. Certain properties and characterizations of a novel family of bivariate 2D-q Hermite polynomials
  111. Stability of an additive-quadratic functional equation in modular spaces
  112. Monotonicity, convexity, and Maclaurin series expansion of Qi's normalized remainder of Maclaurin series expansion with relation to cosine
  113. On k-prime graphs
  114. On the existence of tripartite graphs and n-partite graphs
  115. Classifying pentavalent symmetric graphs of order 12pq
  116. Almost periodic functions on time scales and their properties
  117. Some results on uniqueness and higher order difference equations
  118. Coding of hypersurfaces in Euclidean spaces by a constant vector
  119. Cycle integrals and rational period functions for Γ0+(2) and Γ0+(3)
  120. Degrees of (L, M)-fuzzy bornologies
  121. A matrix approach to determine optimal predictors in a constrained linear mixed model
  122. On ideals of affine semigroups and affine semigroups with maximal embedding dimension
  123. Solutions of linear control systems on Lie groups
  124. A uniqueness result for the fractional Schrödinger-Poisson system with strong singularity
  125. On prime spaces of neutrosophic extended triplet groups
  126. On a generalized Krasnoselskii fixed point theorem
  127. On the relation between one-sided duoness and commutators
  128. Non-homogeneous BVPs for second-order symmetric Hamiltonian systems
  129. Erratum
  130. Erratum to “Infinitesimals via Cauchy sequences: Refining the classical equivalence”
  131. Corrigendum
  132. Corrigendum to “Matrix stretching”
  133. Corrigendum to “A comprehensive review of the recent numerical methods for solving FPDEs”
Downloaded on 9.9.2025 from https://www.degruyterbrill.com/document/doi/10.1515/math-2024-0038/html
Scroll to top button