数学杂志  2019, Vol. 39 Issue (4): 621-632   PDF    
扩展功能
加入收藏夹
复制引文信息
加入引用管理器
Email Alert
RSS
本文作者相关文章
李强
亢婷
陈飞飞
张启敏
带Poisson跳年龄相关随机时滞种群系统的均方稳定性
李强1, 亢婷2, 陈飞飞3, 张启敏4    
1. 东南大学数学学院, 江苏 南京 211189;
2. 宁夏大学新华学院, 宁夏 银川 750021;
3. 南京林业大学林学院, 江苏 南京 210037;
4. 宁夏大学数学与统计学院, 宁夏 银川 750021
摘要:本文研究了带Poisson跳年龄相关随机时滞种群系统均方稳定性的问题.在一定条件下,给出了数值解均方稳定的定义.利用补偿随机θ法讨论系统数值解的均方稳定性,给出数值解稳定的充分条件.获得了当1/2 ≤ θ ≤ 1时,对于任意的步长Δτ/m,数值解是均方稳定的;当0 ≤ θ < 1,时,如果步长Δt∈(0,Δt0),数值解是指数均方稳定的的结果.最后通过数值算例推广并验证了结果的有效性和正确性.
关键词随机时滞种群系统    补偿随机θ    Poisson跳    均方稳定    
MEAN-SQUARE STABILITY OF STOCHASTIC AGE-DEPENDENT DELAY POPULATION SYSTEMS WITH POISSON JUMPS
LI Qiang1, KANG Ting2, CHEN Fei-fei3, ZHANG Qi-min4    
1. School of Mathematics, Southeast University, Nanjing 211189, China;
2. Xinhua College, Ningxia University, Yinchuan 750021, China;
3. School of Forestry, Nanjing Forestry University, Nanjing 210037, China;
4. School of Mathematics and Statistics, Ningxia University, Yinchuan 750021, China
Abstract: This paper deals with the mean-square stability problem of stochastic agedependent delay population systems with Poisson jumps. Under the certain conditions, the definition of mean-square stability of the numerical solution is given. By utilizing the compensated stochastic θ methods, the mean-square stability of the numerical solution is investigated and a sufficient condition for mean-square stability of the numerical solution is presented. It is shown that the compensated stochastic θ methods are mean-square stable for any stepsize Δτ/m when 1/2 ≤ θ ≤ 1, and they are exponentially mean-square stable if the stepsize Δt∈(0, Δt0) when 0 ≤ θ < 1. Finally, the theoretical results are also confirmed by a numerical experiment.
Keywords: stochastic age-dependent delay population systems     compensated stochastic θ method     Poisson jumps     mean-square stability    
1 引言

随机微分方程广泛应用于多个领域, 比如环境, 生物, 金融, 经济和其他科学方面[1-3].近年来, 年龄相关随机种群系统已成为热点研究问题, 并取得了很大的成就.比如Zhang和Liu展现了年龄相关随机种群系统的存在性、唯一性和指数稳定性[4]; Li和Leung研究了带Markov转置年龄相关随机种群系统数值解的收敛性[5]; Ma和Zhang分析了带分数Brown运动年龄相关随机种群系统的数值解[6]; Pang和Li给出了半隐式欧拉法年龄相关随机种群系统数值解的收敛性[7].

然而, 据我们所知, 关于带Poisson跳年龄相关随机时滞种群系统(Stochastic Age-dependent Delay Population Systems, SADDPSs)的均方稳定性研究很少.带跳模型也出现在许多其他应用领域, 并且能描述意想不到的突然的状态变化[8].针对年龄相关随机种群系统, 由于一些突发事件的变化, 比如地球外物体的影响人口系统的规模大大增加或减少, 因此使用跳跃扩散系统能更好得描述人口密度的动态.此外研究这些问题的解的性质是很有价值的.本文我们将讨论如下带Poisson跳年龄相关随机时滞种群系统

$ \begin{align} \left\{ \begin{array}{llll} d_{t}P = -\frac{\partial P}{\partial a}dt-\mu(t, a)Pdt+f(t, P, P_{\tau})dt+g(t, P, P_{\tau})dW_t\\ \ \ \ \ \ \ \ \ \ +h(t, P, P_{\tau})dN_t, &\text{在} \ Q \ \text{内}, \\ P(s, a) = \eta(s, a), & \text{在} \ \overline{R}\ \text{内} , \\ P(t, 0) = \int_0^{A}\beta(t, a)P(t, a) da, & \text{在} \ [0, T]\ \text{上}, \end{array} \right. \end{align} $ (1)

其中$ a\in[0, A] $为年龄, $ t\in[0, T] $为时间, $ P = P(t, a) $, $ Q = (0, T)\times(0, A) $, $ \overline{R} = [-\tau, 0]\times[0, A] $, $ P_0 = P(0, a) = \eta(0, a) $, $ \tau $为时滞, $ P_{\tau} = P(t-\tau, a) $.时滞意味着种群密度和以前某时刻的种群密度是相互独立的. $ d_{t}P $表示$ P $$ t $的微分, 即$ d_{t}P $ = $ \frac{\partial P}{\partial t}dt $, $ P(t, a) $表示在$ t $时刻年龄为$ a $的种群密度, $ \beta(t, a) $表示在$ t $时刻年龄为$ a $的种群的生育率, $ \mu(t, a) $表示在$ t $时刻年龄为$ a $的种群死亡率, $ W_t $是一个标准的Brown运动, $ f(t, P, P_{\tau})dt+g(t, P, P_{\tau})dW_t+h(t, P, P_{\tau})dN_t $表示外界环境对种群系统的扰动, 比如迁移、地震、火山爆等等.

本文对带Poisson跳年龄相关随机时滞种群系统的均方稳定性问题讨论是对文献[9, 15]的扩展, 具有重要的研究意义.

论文安排如下, 我们研究了带Poisson跳年龄相关随机种群系统补偿随机$ \theta $法的均方稳定性.第二部分给出了一些定义, 并且也给出了数值解均方稳定性的定义和条件; 第三部分讨论了系统解析解的均方稳定性; 第四部分研究了补偿随机$ \theta $法的均方稳定性; 第五部分, 给出了一个数值算例对结论的有效性和正确性进行了验证.

2 预备知识

$ V = H^1([0, A])\equiv \{\varphi | \varphi\in L^2([0, A]), \frac{\partial \varphi}{\partial a}\in L^2([0, A]), \mbox {其中}\; \frac{\partial \varphi}{\partial a} \; \mbox{是广义偏导数}\}, $ $ V $是Sobolev空间, $ H = L^2([0, A]) $, 满足

$ V\hookrightarrow H\equiv H'\hookrightarrow V'. $

$ V' = H^{-1}([0, A]) $$ V $的对偶空间; 定义$ |\cdot| $$ \|\cdot\| $分别为$ V $, $ V' $上的范数; $ \langle \cdot, \cdot\rangle $表示$ V $$ V' $空间的内积; $ (\cdot, \cdot) $$ H $空间上的数量积. $ m $是一个常数, 即有

$ |x|\leq m\|x \|, \ \ \ \ \forall x\in V. $

$ \mathcal {C} = \mathcal {C}([-\tau, 0]; H) $表示所有从$ [-\tau, 0] $$ H $的连续函数组成的空间, 其范数定义为$ \|\phi\|: = \sup\limits_{-\tau\leq s\leq 0}|\phi(s)| $, $ L^p_V = L^p([-\tau, 0]; V) $$ L_H^p = L^p([-\tau, 0]; H) $. $ \mathscr{W}([-\tau, 0];\overline{R}_{+}) $是Borel可测集中的非负函数, 满足$ \sup\limits_{-\tau\leq s\leq 0}\mid \eta(s)\mid\leq \xi<\infty $, $ \xi \in \mathscr{W}([-\tau, 0];\overline{R}_{+}) $.

$ (\Omega, {\cal F}, \{{\cal F}_{t}\}_{t\geq 0}, \mathbb{P}) $是完备概率空间, 假设$ \{{\cal F}_{t}\}_{t\geq 0} $是由三个完全独立的过程生成的滤波算子(左极限是右连续的, 并且$ {\cal F}_{0} $包含所有的零测集). $ W_{t} $是定义在完备的概率空间$ (\Omega, {\cal F}, P) $上并且取值在可分的Hilbert空间$ K $上的Wiener过程, 具有增量协方差算子$ W $. $ N_{t} $是定义在相同的概率空间带有参数$ \lambda $的标准Poisson过程.设$ W_{t} $$ N_{t} $是相互独立的. $ B\in \mathcal {L}(K, H) $是所有从$ K $$ H $的有界线性算子空间, $ \|B\|_2 $表示Hilbert-Schmidt范数, 即

$ \|B\|^2_2 = {\hbox{tr}}(BWB^T). $

并且假设$ \phi(t) $$ {\cal F}_0 $可测的, 右连续的, 有$ E\|\phi\|^2<\infty $.我们也假设$ f(t, 0, 0) = 0 $, $ g(t, 0, 0) = 0 $, $ h(t, 0, 0) = 0 $, 这样系统(1)有零解$ P(t, a)\equiv 0 $.

定义 2.1   (见文[16])如果存在一系列正常数$ \lambda $$ C $, 使得

$ \begin{align} E|P(t)|^p\leq C\|\phi\|{e}^{-\lambda(t-t_0)}, \; \; \; \; t\geq t_0, \end{align} $ (2)

那么系统(1)的零解是$ p $界矩指数稳定.当$ p = 2 $时, 通常为指数均方稳定.

为了证明本章的主要结论, 我们给出如下假设条件.

(i)   $ \mu(t, a) $, $ \beta(t, a) $$ V $上非负可测的, 存在常数$ \mu_0 $$ \bar{\beta} $, 使得

$ \left\{ \begin{array}{ll} 0\leq\mu(t, a)\leq\mu_0<\infty, &\text{在} \ Q \ \text{内}, \\ 0\leq\beta(t, a)\leq\overline{\beta}\leq\infty, &\text{在} \ Q \ \text{内}, \\ A\overline{\beta}-2\mu_0<0, &\text{在} \ Q \ \text{内}. \end{array} \right. $

(ii)   存在一些常数$ b_i $, $ c_i $, $ d_i $, $ i = 1, 2 $满足

$ \begin{align} \langle x_1-x_2, f(t, x_1, y)-f(t, x_2, y)\rangle\leq b_1|x_1-x_2|^2, \end{align} $ (3)
$ \begin{align} |f(t, x, y_1)-f(t, x, y_2)|\leq b_2|y_1-y_2|, \end{align} $ (4)
$ \begin{align} |g(t, x_1, y_1)-g(t, x_2, y_2)|^2\leq c_1|x_1-x_2|^2+c_2|y_1-y_2|^2 , \end{align} $ (5)
$ \begin{align} |h(t, x_1, y_1)-h(t, x_2, y_2)|^2\leq d_1|x_1-x_2|^2+d_2|y_1-y_2|^2. \end{align} $ (6)
3 带Poisson跳SADDPSs(1)的均方稳定性

下面给出带Poisson跳SADDPSs (1)具有均方稳定性的一些充分条件.

定理 3.1  假设系统(1)满足条件(i)–(ii), $ P(t): = P(t, a) $是其解, 令$ \alpha\equiv2b_1+2b_2+c_1+c_2+\lambda(1+2d_1+2d_2)<0, $那么系统(1)的解析解是指数均方稳定的.

  令$ t\geq 0, $ $ \delta> 0, $利用Itô公式, 得

$ \begin{align} \begin{array}{llll} |P(t+\delta)|^2 = &|P(t)|^2+ \int_t^{t+\delta}(2\langle P(s), -\frac{\partial P(s)}{\partial a}-\mu(s, a)P(s)+f(s, P(s), P(s-\tau))\rangle)ds\\ &+2 \int_t^{t+\delta}\langle P(s), g(s, P(s), P(s-\tau))\rangle dW(s) + \int_t^{t+\delta}|g(s, P(s), P(s-\tau))|^2ds\\ &+ \int_t^{t+\delta}(2\langle P(s), h(s, P(s), P(s-\tau))\rangle+|h(s, P(s), P(s-\tau))|^2)d\tilde{N}(s)\\ &+\lambda \int_t^{t+\delta}(2\langle P(s), h(s, P(s), P(s-\tau))\rangle+|h(s, P(s), P(s-\tau))|^2)ds, \end{array} \end{align} $ (7)

其中$ \tilde{N}(t) = N(t)-\lambda t $是一个补偿的Poisson过程.等式(7)的两边取期望并用Itô积分性质可以得到

$ \begin{align} \begin{array}{llll} E|P(t+\delta)|^2 = &E|P(t)|^2+ \int_t^{t+\delta}E(2\langle P(s), -\frac{\partial P(s)}{\partial a}-\mu(s, a)P(s)\\ &+f(s, P(s), P(s-\tau))\rangle)ds+E \int_t^{t+\delta}|g(s, P(s), P(s-\tau))|^2ds\\ &+E\lambda \int_t^{t+\delta}(2\langle P(s), h(s, P(s), P(s-\tau))\rangle+|h(s, P(s), P(s-\tau))|^2)ds. \end{array} \end{align} $ (8)

从式$ (3) $$ (4) $, 可以得到

$ \begin{align} \begin{array}{ll} &2E\langle P(s), f(s, P(s), P(s-\tau)\rangle\\ = &2E\langle P(s), f(s, P(s), P(s-\tau)-f(0, 0, P(s-\tau)\rangle +2E\langle (P(s), f(0, 0, P(s-\tau)\rangle \\ \leq& 2b_1E|P(s)|^2+2b_2E|P(s)||P(s-\tau)| \leq 2b_1E|P(s)|^2+2b_2\sup\limits_{s-\tau\leq u\leq s}E|P(u)|^2. \end{array} \end{align} $ (9)

相似地, 从式$ (5) $$ (6) $可以得到

$ \begin{align} \begin{array}{ll} &E|g(s, P(s), P(s-\tau)|^2\leq(c_1+c_2)\sup\limits_{s-\tau\leq u\leq s}E|P(u)|^2, \\ &E|h(s, P(s), P(s-\tau)|^2\leq(d_1+d_2)\sup\limits_{s-\tau\leq u\leq s}E|P(u)|^2, \\ &2E\langle P(s), h(s, P(s), P(s-\tau)\rangle\leq(1+c_1+c_2)\sup\limits_{s-\tau\leq u\leq s}E|P(u)|^2. \end{array} \end{align} $ (10)

由于

$ \begin{align} \begin{array}{ll} -\langle\frac{\partial P(s)}{\partial a}, P(s)\rangle& = - \int_0^A P(s)d_{a}(P(s)) = \frac{1}{2}( \int_0^A\beta(s, a)P(s)da)^2\nonumber\\ &\leq \frac{1}{2} \int_0^A\beta^2(s, a)da \int_0^A|P(s)|^2da \leq\frac{1}{2}A\bar{\beta}^2|P(s)|^2. \end{array} \end{align} $ (11)

把式(9)–(11)和假设条件(i)带入到式(8)中, 可以得出

$ \begin{align} \begin{array}{ll} E|P(t+\delta)|^2&\leq E|P(t)|^2+ \int_t^{t+\delta}((A\bar{\beta}-2\mu_0)E|P(s)|^2+2b_1E|P(s)|^2+(2b_2+c_1+c_2\\ &\ \ \ \ +\lambda(1+2d_1+2d_2))\sup\limits_{s-\tau\leq u\leq s}E|P(u)|^2)ds\\ &\leq E|P(t)|^2+ \int_t^{t+\delta}(2b_1E|P(s)|^2+(2b_2+c_1+c_2\\ &\ \ \ \ +\lambda(1+2d_1+2d_2))\sup\limits_{s-\tau\leq u\leq s}E|P(u)|^2)ds. \end{array} \end{align} $ (12)

$ \nu(t) = E|P(t)|^2 $, $ \varepsilon = 2b_2+c_1+c_2+\lambda(1+2d_1+2d_2) $, 则

$ \begin{align} D^{+}\nu(t)\leq 2b_1\nu(t)+\varepsilon\sup\limits_{t-\tau\leq u\leq t}\nu(u), \end{align} $ (13)

其中

$ \begin{align} D^{+}\nu(t) = \lim\sup\limits_{\delta\searrow0} \frac{\nu(t+\delta)-\nu(t)}{\delta}. \end{align} $ (14)

由于$ \alpha<0 $, 则$ -2b_1>\varepsilon\geq 0 $.结合文献[4]中引理1.1, 存在正常数$ \kappa $$ \sigma $使得

$ \nu(t)\leq\kappa e^{-\sigma t}, \ \ \ \ \ t\geq0. $

基于上述结果, 以下部分我们将研究系统(1)数值解补偿随机$ \theta $法的稳定性.

4 带Poisson跳SADDPSs(1)补偿随机$ \theta $法的均方稳定性

这一部分, 将研究系统(1)补偿随机$ \theta $法的均方稳定性.由于补偿的Poisson过程$ \tilde{N(t)} = N(t)-\lambda t $是一个鞅, 满足以下的性质

$ \begin{align} E(\tilde{N}_{(t+s)}-\tilde{N}_t) = 0, \ \ \ \ \; \; \; \; \; \; \; E|\tilde{N}_{(t+s)}-\tilde{N}_t)|^2 = \lambda s, \; \; \; \; t, s\geq0. \end{align} $ (15)

重新写系统(1)的等价形式

$ \begin{align} d_{t}P = -\frac{\partial P}{\partial a}dt-\mu(t, a)Pdt+\tilde{f}(t, P, P_{\tau})dt+g(t, P, P_{\tau})dW_t +h(t, P, P_{\tau})d\tilde{N}_t, \end{align} $ (16)

其中$ \tilde{f}(t, P, P_{\tau}) $被定义为

$ \begin{align} \tilde{f}(t, P, P_{\tau}): = f(t, P, P_{\tau})+\lambda h(t, P, P_{\tau}). \end{align} $ (17)

对式(17)利用随机$ \theta $方法诱导出以下补偿随机$ \theta $方法的形式

$ \begin{align} \begin{array}{ll} P_{n+1} = &P_n+(1-\theta)(-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m}))\Delta t\\&+\theta(-\frac{\partial P_{n+1}}{\partial a} -\mu(t_{n+1}, a)P_{n+1} +\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1}))\Delta t\\&+g(t_n, P_n, P_{n-m})\Delta W_n+h(t_n, P_n, P_{n-m})\Delta \tilde{N}_n. \end{array} \end{align} $ (18)

在此, $ P_n $表示$ P(t, a) $的逼近, 参数$ \theta $的范围为$ 0\leq \theta\leq 1 $, $ \Delta t: = t_{n+1}-t_{n} $是步长, 存在正整数$ m $满足$ \tau = m\Delta t $, 正整数$ N $满足$ T = N\Delta t $, $ W_n: = W_{t_{n+1}}-W_{t_{n}} $$ \Delta\tilde{N}_n: = \tilde{N}_{t_{n+1}}-\tilde{N}_{t_{n}} $.

定义 4.1  给出步长$ \Delta t = \tau/m $, 如果带Poisson跳SADDPSs (1)的数值解$ P_n $满足

$ \begin{align} \lim\limits_{n\longrightarrow \infty}E|P_{n}|^{2} = 0. \end{align} $ (19)

则系统(1)的数值解$ P_n $是均方稳定的.

定理 4.1  假设条件(3)–(6)成立.如果$ 1/2\leq\theta\leq1 $, 对于每个步长$ \Delta t = \tau/m $补偿随机$ \theta $法是均方稳定的.

  从式(18)可以得到

$ \begin{align} \begin{array}{ll} &|P_{n+1}-\theta\Delta t(-\frac{\partial P_{n+1}}{\partial a} -\mu(t_{n+1}, a)P_{n+1}+\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1}))|^2\\ = &|P_n-\theta(-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m}))\Delta t|^2\\ &+2\Delta t\langle P_n, -\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m})\rangle\\ &+(\Delta t)^2(1-2\theta)|-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m})|^2\\ &+|g(t_n, P_n, P_{n-m})\Delta W_n|^2+|h(t_n, P_n, P_{n-m})\Delta \tilde{N}_n|^2+M_n, \end{array} \end{align} $ (20)

其中

$ \begin{align} \begin{array}{ll} M_n = &2\langle P_n+(1-\theta)(-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m})), g(t_n, P_n, P_{n-m})\Delta W_n\rangle\\ &+2\langle P_n+(1-\theta)(-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m})), h(t_n, P_n, P_{n-m})\Delta \tilde{N}_n\rangle\\ &+2\langle g(t_n, P_n, P_{n-m})\Delta W_n, h(t_n, P_n, P_{n-m})\Delta \tilde{N}_n\rangle. \end{array} \end{align} $ (21)

这样对于$ 1/2\leq\theta\leq1 $, 有

$ \begin{align} \begin{array}{ll} &|P_{n+1}-\theta\Delta t(-\frac{\partial P_{n+1}}{\partial a} -\mu(t_{n+1}, a)P_{n+1}+\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1}))|^2\\ \leq&|P_n-\theta(-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m}))\Delta t|^2\\ &+2\Delta t\langle P_n, -\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m})\rangle\\ &+|g(t_n, P_n, P_{n-m})\Delta W_n|^2+|h(t_n, P_n, P_{n-m})\Delta \tilde{N}_n|^2+M_n.\\ \end{array} \end{align} $ (22)

从式(3), (4)和(6), 可以得到

$ \begin{align} \begin{array}{ll} &2\langle P_n, \tilde{f}(t_n, P_n, P_{n-m})\rangle\\ = &2\langle P_n, f(t_n, P_n, P_{n-m})\rangle+2\lambda\langle P_n, h(t_n, P_n, P_{n-m})\rangle\\ \leq&2b_1|P_n|^2+b_2|P_n|^2+b_2|P_{n-m}|^2 +\lambda(|P_n|^2+d_1|P_n|^2+d_2|P_{n-m}|^2) \end{array} \end{align} $ (23)

$ \begin{align} \begin{array}{ll} -\langle\frac{\partial P_{n}}{\partial a}, P_{n}\rangle& = - \int_0^A P_{n}d_{a}(P_{n}) = \frac{1}{2}( \int_0^A\beta(t_n, a)P_{n}da)^2\nonumber\\ &\leq \frac{1}{2} \int_0^A\beta^2(t_n, a)da \int_0^A|P_{n}|^2da \leq\frac{1}{2}A\bar{\beta}^2|P_{n}|^2. \end{array} \end{align} $ (24)

由于$ E(\Delta W_n) = 0 $, $ E(\Delta\tilde{N}_n) = 0 $$ E(\Delta\tilde{N}_n)^2 = \lambda\Delta t $,且$ P_n $$ P_{n-m} $都是$ {\mathcal {F}_{t_n}} $可测的.因此很容易得到

$ \begin{align} E|g(t_n, P_n, P_{n-m})\Delta W_n|^2 = \Delta tE|g(t_n, P_n, P_{n-m})|^2, \end{align} $ (25)
$ \begin{align} E|h(t_n, P_n, P_{n-m})\Delta \tilde{N}_n|^2 = \lambda\Delta tE|h(t_n, P_n, P_{n-m})|^2, \end{align} $ (26)
$ \begin{align} EM_n = 0. \end{align} $ (27)

对式(22)两边取期望, 把式(23)–(27)带入到式(22)中, 有

$ \begin{align} \begin{array}{ll} &E|P_{n+1}-\theta\Delta t(-\frac{\partial P_{n+1}}{\partial a} -\mu(t_{n+1}, a)P_{n+1}+\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1}))|^2\\ \leq& E |P_n-\theta(-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m}))\Delta t|^2\\ &+\Delta t(A\bar{\beta}^2-2\mu_0+2b_1+b_2+\lambda+c_1+2\lambda d_1)E|P_n|^2\\ &+\Delta t(b_2+c_2+2\lambda d_2)E|P_{n-m}|^2. \end{array} \end{align} $ (28)

因此对不等式(28)递归计算得到

$ \begin{align} \begin{array}{ll} &E|P_{n+1}-\theta\Delta t(-\frac{\partial P_{n+1}}{\partial a} -\mu(t_{n+1}, a)P_{n+1}+\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1}))|^2\\ \leq& E |P_n-\theta(-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m}))\Delta t|^2\\ &+\Delta t(A\bar{\beta}^2-2\mu_0+2b_1+b_2+\lambda+c_1+2\lambda d_1)E|P_n|^2\\ &+\Delta t(b_2+c_2+2\lambda d_2)E|P_{n-m}|^2\\ \leq& E|P_{n-1}-\theta\Delta t(-\frac{\partial P_{n-1}}{\partial a}-\mu(t_{n-1}, a)P_{n-1}+\tilde{f}(t_{n-1}, P_{n-1}, P_{n-m-1}))|^2\\ &+\Delta t(A\bar{\beta}^2-2\mu_0+2b_1+b_2+\lambda+c_1+2\lambda d_1)\sum\limits_{j = n-1}^n E|P_j|^2\\ &+\Delta t(b_2+c_2+2\lambda d_2)\sum\limits_{j = n-1}^n E|P_{j-m}|^2\\ \leq& \vdots\\ \leq& E |P_0-\theta(-\frac{\partial P_0}{\partial a}-\mu(t_0, a)P_0+\tilde{f}(t_0, P_0, P_{-m}))\Delta t|^2\\ &+\Delta t(A\bar{\beta}^2-2\mu_0+2b_1+b_2+\lambda+c_1+2\lambda d_1)\sum\limits_{j = 0}^n E|P_j|^2\\ &+\Delta t(b_2+c_2+2\lambda d_2)\sum\limits_{j = 0}^n E|P_{j-m}|^2. \end{array} \end{align} $ (29)

标记$ \sum\limits_{j = 0}^n E|P_{j-m}|^2 = \sum\limits_{j = -m}^{-1} E|P_{j}|^2+\sum\limits_{j = 0}^{n-m} E|P_{j}|^2 $$ \tau = m\Delta t $, 从式(29)可得到

$ \begin{align} \begin{array}{ll} \; \; \; \; &E|P_{n+1}-\theta\Delta t(-\frac{ \partial P_{n+1}}{\partial a} -\mu(t_{n+1}, a)P_{n+1}+\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1}))|^2\\ \leq& E |P_0-\theta(-\frac{\partial P_0}{ \partial a}-\mu(t_0, a)P_0+\tilde{f}(t_0, P_0, P_{-m}))\Delta t|^2\\ &+\Delta t(A\bar{\beta}^2-2\mu_0+2b_1+2b_2+c_1+c_2+\lambda(1+2d_1+2d_2))\sum\limits_{j = 0}^n E|P_j|^2\\ &+\tau(b_2+c_2+2\lambda d_2)\times\max\limits_{-m\leq j\leq 0} E|P_{j}|^2. \end{array} \end{align} $ (30)

重新整理式(30), 利用假设条件(i), 得到

$ \begin{align} \begin{array}{ll} \; \; \; \; & E |P_{n+1}-\theta\Delta t(-\frac{\partial P_{n+1}}{\partial a} -\mu(t_{n+1}, a)P_{n+1}\\ &+\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1}))|^2 -\alpha\Delta t\sum\limits_{j = 0}^n E|P_j|^2\\ \leq& E |P_0-\theta(-\frac{\partial P_0}{\partial a}-\mu(t_0, a)P_0+\tilde{f}(t_0, P_0, P_{-m}))\Delta t|^2\\ &+\tau(b_2+c_2+2\lambda d_2)\times\max\limits_{-m\leq j\leq 0} E|P_{j}|^2. \end{array} \end{align} $ (31)

由于$ E{\|\phi \|}^2<\infty $$ \alpha < 0 $, 可以推算出级数$ \sum\limits_{j = 0}^{\infty} E|P_{j}|^2 $是收敛的, 可以得出$ \lim\limits_{n\rightarrow\infty}E|P_{n}|^2 = 0 $.因此对于$ 1/2\leq\theta\leq1 $, 对于每个步长$ \Delta t = \tau/m $补偿随机$ \theta $法是均方稳定的.

为了给出数值方法的指数稳定性, 需要给出以下引理, 在文献[17]中是定理1.

引理 4.1   (见文[17])给出一些确定的整数$ N\geq 0 $, 对于$ \Delta t>0 $, 假设$ t_n = t_0+n\Delta t $, $ \{\upsilon_n\}_{-N}^{\infty} $是一系列正数序列并满足

$ \begin{align} \frac{\upsilon_{n+1}-\upsilon_{n}}{\Delta t}\leq -\alpha_{\Delta t}\upsilon_{n}+\beta_{\Delta t}\max\limits_{j\in\mathscr{T}}\upsilon_{n+j} , \; \; \; n\in\mathscr{N}. \end{align} $ (32)

如果$ \beta_{\Delta t} = 0, $$ N = 0 $, 其中$ \mathscr{T} : = \{-N, \cdots, -1, 0\} $.如果

$ \begin{align} 0\leq\beta_{\Delta t}<\alpha_{\Delta t}, \; \; \; \; 0<\alpha_{\Delta t}\Delta t<1, \end{align} $ (33)

然后$ \upsilon_n\leq\{\max\limits_{j\in\mathscr{T}}\upsilon_{j}\}\exp\{-\upsilon^{+}(t_n-t_0)\}, $其中$ \upsilon^{+}>0 $是一个常数.

定理 4.2  假设条件(3)—(6)成立, 漂移系数$ f $满足线性增长条件, 存在一个常数$ D $使得

$ \begin{align} |f(t, x, y)|^2\leq D(|x|^2+|y|^2). \end{align} $ (34)

标记

$ \begin{eqnarray*} &&\Delta t_1 = -\alpha/2(1-\theta)^2(2D+{\lambda}^2(d_1+d_2)), \\ &&\Delta t_2 = -(2b_1+b_2+c_1+\lambda+2\lambda c_1)/2(1-\theta)^2(D+{\lambda}^2d_1), \\ &&\Delta t_3 = \inf\{\Delta t>0:N(\theta, \Delta t)<0\}, \end{eqnarray*} $

其中

$ N(\theta, \Delta t) = 2(1-\theta)^2(D+{\lambda}^2d_1)(\Delta t)^2+(1-\theta)(2b_1+b_2+\lambda+\lambda c_1)+c_1+\lambda d_1)\Delta t+1. $

如果$ 0\leq\theta<1 $和步长$ \Delta t\in (0, \Delta t_0) $并且$ \Delta t_0 = \min\{\Delta t_1, \Delta t_2, \Delta t_3\}, $补偿随机$ \theta $方法是指数均方稳定的.

  从式(18)可以得到

$ \begin{align} \begin{array}{ll} &|P_{n+1}-\theta(-\frac{\partial P_{n+1}}{\partial a}-\mu(t_{n+1}, a)P_{n+1}-\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1}))\Delta t|^2\\ = &|P_n+(1-\theta)(-\frac{\partial P_n}{\partial a}-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m}))\Delta t\\ &+g(t_n, P_n, P_{n-m})\Delta W_n+h(t_n, P_n, P_{n-m})\Delta \tilde{N}_n|^2. \end{array} \end{align} $ (35)

因此有

$ \begin{align} \begin{array}{ll} |P_{n+1}|^2\leq&|P_{n}|^2+(1-\theta)^2(\Delta t)^2|\tilde{f}(t_n, P_n, P_{n-m})|^2+|g(t_n, P_n, P_{n-m})\Delta W_n|^2\\ &+|h(t_n, P_n, P_{n-m})\Delta \tilde{N}_n|^2+2\theta\Delta t\langle P_{n+1}, -\frac{\partial P_{n+1}}{\partial a}-\mu(t_{n+1}, a)P_{n+1}\\ &-\tilde{f}(t_{n+1}, P_{n+1}, P_{n-m+1})\rangle+2(1-\theta)\langle P_n, -\frac{\partial P_n}{\partial a}\\ &-\mu(t_n, a)P_n+\tilde{f}(t_n, P_n, P_{n-m})\rangle+M_n, \end{array} \end{align} $ (36)

其中$ M_n $定义在式(21).从式(6), (17)和(34), 可以得到

$ \begin{align} \begin{array}{ll} |\tilde{f}(t_n, P_n, P_{n-m})|^2& = |f(t_n, P_n, P_{n-m})+\lambda h(t_n, P_n, P_{n-m})|^2\\ &\leq 2(D(|P_n|^2+|P_{n-m}|^2)+{\lambda}^2(d_1|P_n|^2+d_2|P_{n-m}|^2)). \end{array} \end{align} $ (37)

将式(23)–(27)和式(37)带入到式(36)中, 取期望可以得到

$ \begin{align} \begin{array}{ll} E|P_{n+1}|^2\leq& E|P_{n}|^2+2(1-\theta)^2(\Delta t)^2((D+{\lambda}^2d_1)E|P_{n}|^2\\ &+(D+{\lambda}^2d_2)E|P_{n-m}|^2)+\Delta t(c_1E|P_{n}|^2+c_2E|P_{n-m}|^2)\\ &+\lambda\Delta t(d_1E|P_{n}|^2+d_2E|P_{n-m}|^2)+\theta\Delta t((2b_1+b_2\\ &+\lambda(1+d_1))E|P_{n+1}|^2+(b_2+\lambda d_2)E|P_{n-m+1}|^2+(1-\theta)\Delta t((2b_1+b_2\\ &+\lambda(1+d_1))E|P_{n}|^2+(b_2+\lambda d_2)E|P_{n-m}|^2, \end{array} \end{align} $ (38)

可以导出

$ \begin{align} \begin{array}{ll} &(1-\theta\Delta t((2b_1+b_2+\lambda(1+d_1))E|P_{n+1}|^2\\ \leq&(1-\theta\Delta t((2b_1+b_2+\lambda(1+d_1))E|P_{n}|^2+((2b_1+b_2+c_1+\lambda+2\lambda d_1)\Delta t\\ &+2(1-\theta)^2(D+{\lambda}^2d_1)(\Delta t)^2)E|P_{n}|^2\\ &+((b_2+c_2+2\lambda d_2)\Delta t+2(1-\theta)^2(D+{\lambda}^2d_2)(\Delta t)^2)\max\limits_{n-m\leq i\leq n-m+1}E|P_i|^2. \end{array} \end{align} $ (39)

因此

$ \begin{align} \frac{E|P_{n+1}|^2-E|P_{n}|^2}{\Delta t}\leq-A E|P_{n}|^2+B \max\limits_{n-m\leq i\leq n-m+1}E|P_i|^2, \end{align} $ (40)

其中

$ \begin{align} A = -\frac{2b_1+b_2+c_1+\lambda+2\lambda d_1+2(1-\theta)^2(D+{\lambda}^2d_1)\Delta t}{1-\theta\Delta t(2b_1+b_2+\lambda(1+d_1))}, \end{align} $ (41)
$ \begin{align} B = \frac{b_2+c_2+2\lambda d_2+2(1-\theta)^2(D+{\lambda}^2d_2)\Delta t}{1-\theta\Delta t(2b_1+b_2+\lambda(1+d_1))}. \end{align} $ (42)

结合引理4.1, 得出步长随机$ \theta $方法是均方稳定的, 如果

$ \begin{align} 0\leq B<A, \; \; \; 0<A\Delta t<1. \end{align} $ (43)

可以得出

$ \begin{align} \begin{array}{ll} B-A = \frac{2b_1+2b_2+c_1+c_2+\lambda(1+2d_1+2d_2)+2(1-\theta)^2(2D+{\lambda}^2(d_1+d_2))\Delta t}{1-\theta\Delta t(2b_1+b_2+\lambda(1+d_1))}<0 \end{array} \end{align} $ (44)

$ \begin{align} \begin{array}{ll} A = -\frac{2b_1+b_2+c_1+\lambda+2\lambda d_1+2(1-\theta)^2(D+{\lambda}^2d_1)\Delta t}{1-\theta\Delta t(2b_1+b_2+\lambda(1+d_1))}>0. \end{array} \end{align} $ (45)

明显地

$ \begin{eqnarray*} &&\Delta t<\frac{-\alpha}{2(1-\theta)^2(2D+{\lambda}^2(d_1+d_2)}, \\ &&\Delta t<-\frac{2b_1+b_2+c_1+\lambda+2\lambda d_1}{2(1-\theta)^2(D+{\lambda}^2d_1)}, \; \; N(\theta, \Delta t)>0, \end{eqnarray*} $

其中

$ N(\theta, \Delta t) = 2(1-\theta)^2(D+{\lambda}^2d_1)(\Delta t)^2 +((1-\theta)(2b_1+b_2+\lambda+\lambda d_1)+c_1+\lambda d_1){\Delta} t+1. $

因为$ N(\theta, 0) = 1, $则必存在$ \Delta t_3>0 $, 使得当$ \Delta t<\Delta t_3 $时, $ N(\theta, \Delta t)>0 $.另一方面, 如果$ N(\theta, \Delta t)\geq0 $总是正确的, 定义$ \Delta t_3 $$ \infty. $因此令

$ \begin{eqnarray*} &&\Delta t_1 = -\alpha/2(1-\theta)^2(2D+{\lambda}^2(d_1+d_2)), \\&&\Delta t_2 = -(2b_1+b_2+c_1+\lambda+2\lambda d_1)/2(1-\theta)^2(D+{\lambda}^2d_1), \\ &&\Delta t_3 = \inf\{\Delta t>0:N(\theta, \Delta t)<0\}, \\ &&\Delta t_0 = \min\{\Delta t_1, \Delta t_2, \Delta t_3\}, \end{eqnarray*} $

$ \Delta t\in(0, \Delta t_0) $时, 式(43)成立, 即定理4.2得证.

通过证明定理4.2, 很容易能得到以下的结果.

定理 4.3  假设(3)–(6)式和$ \alpha<0 $成立, 如果$ \theta = 1 $, 对于每个步长$ \Delta t = \tau/m $, 补偿随机$ \theta $法是指数均方稳定的.

5 数值算例

这一部分的主要目的是通过数值算例验证结论的有效性和正确性.考虑如下带Poisson跳年龄相关随机时滞种群系统.

$ \begin{align} \left\{ \begin{array}{ll} d _tP = [-\frac{\partial P}{\partial a} -\frac{1}{(1-a)^2}P-t\times P\times P_{\tau}]dt+(P_{\tau}+1)\times P dW_t\\ \ \ \ \ \ \ \ \ +(P_{\tau}+1)\times P dN_t, &\text{在} \ Q \ \text{内}, \\ P(s, a) = \exp(-\frac{0.5+s}{1-a}), & \text{在} \ \ \overline{R} \ \text{内}, \\ P(t, 0) = \int_0^{1}-\frac{1}{(1-a)^2}P(t, a) da, &\text{在} \ \ [0, T] \text{内}, \end{array} \right. \end{align} $ (46)

其中$ Q = (0, T)\times (0, 1) $, $ \overline{R} = [-\tau, 0]\times[0, 1] $, $ P_{0} = P(0, a) = \eta(0, a) $, $ a = 5 $, $ P_\tau = P(t-\tau, a) $, $ \tau = 0.5 $是时滞, 满足

$ \begin{eqnarray*} &&\sup\limits_{-\tau\leq s\leq 0}\mid \eta(s)\mid\leq 1, \; \; \mu(t, a) = \beta(t, a) = \frac{1}{{(1-a)^2}}, \; \; f(t, P, P_{\tau}) = -t\times P\times P_{\tau}, \\ &&g(t, P, P_{\tau}) = h(t, P, P_{\tau}) = (P_{\tau}+1)\times P, \; \; \eta(s, a) = \exp(-\frac{0.5+s}{1-a}). \end{eqnarray*} $

$ W_t $是标准的Brownian运动, $ N_t $是带有参数$ \lambda = 1 $标准的Poisson过程.在这种情况下, 条件(3)–(6)和$ \alpha<0 $是满足的. $ b_1 = -4, b_2 = 1, c_1 = 0, c_2 = 1, d_1 = 1 $$ d_2 = 0 $, 这样可以得到$ \alpha = -2 $, 确保系统(46)在给出的条件下是均方稳定的.

接下来的数值模拟将会呈现出补偿随机$ \theta $法中的参数$ \theta $和步长$ \Delta t $是如何影响系统数值解的均方稳定性.我们对$ E|P_n|^2 $模拟1000次, 也就是

$ E|P_n|^2 = \frac{1}{1000}\sum\limits_{i = 1}^{1000}|P_i|^2. $

定理4.1说明在假设条件(3)–(6)成立下.当$ 1/2\leq\theta\leq1 $, 对于每个步长$ \Delta t = \tau/m $补偿随机$ \theta $法是均方稳定的. 图 1是分别选取$ \theta $为0.5和0.8, $ \Delta t $$ 1/8 $, $ 1/4 $, $ 1/2 $, $ 1 $利用式(18)解决(46)得到的.从图 1发现在这些步长下补偿随机$ \theta $法是均方稳定的.

图 1 (a) 当$\theta = 0.5$时; (b)当$\theta = 0.8$时, 在步长为$\Delta t = 1/8, 1/4, 1/2, 1$数值解的轨迹

定理4.2说明在假设条件(3)–(6)成立下, 当$ 0\leq\theta\leq1 $, 对于步长$ \Delta t\in (0, \Delta t_0) $补偿随机$ \theta $法是指数均方稳定的. $ N_t $是带有参数$ \lambda = 1 $的标准的Poisson过程.式(34)成立, $ b_1 = -3, b_2 = 0, c_1 = 0, c_2 = 1, d_1 = 0, d_2 = 1, \tau = 1, D = 9 $$ \alpha = -2 $, 因此系统(46)是指数均方稳定的.根据定理4.2, 计算出$ \Delta t_1 = 1/19(1-\theta)^2, \Delta t_2 = 5/18(1-\theta)^2, $$ N(\theta, \Delta t) = 18(1-\theta)^2(\Delta t)^2-5(1-\theta)\Delta t+1 $.很明显得到对于每个步长$ \Delta t>0 $, $ N(\theta, \Delta t)>0 $, 取$ \Delta t_3 = \infty $, 因此得到$ \Delta t_0 = \min\{\Delta t_1, \Delta t_2, \Delta t_3\} $, 将补偿随机$ \theta $法用到系统(46), 随着参数$ \theta $的增加, 步长限制很小. 图 2是我们分别选取$ \theta $为0和0.2, $ \Delta t $$ 1/20 $, $ 1/8 $, $ 1/2 $, $ 1 $利用(18)时候解决(46)式得到的.结合定理4.2, 我们计算出$ \theta = 0 $时, $ \Delta t_0 = 0.0526 $$ \theta = 0.2 $时, $ \Delta t_0 = 0.0822 $.从图 2发现在步长$ \Delta t = 0.05 $下, 补偿随机$ \theta $法是指数均方稳定的, $ \Delta t $最好在$ (0, \Delta t_0) $中选取.当步长选取大于$ \Delta t_0 $时, 这种方法将不会稳定. $ \Delta t = 0.125 $时, 这种方法稳定, $ \Delta t = 1 $时, 这种方法将会不稳定, 这表明在定理4.2中, 步长$ \Delta t_0 $的限制不是最优的.并且从图 2中也发现参数$ \theta $取值越大, 步长$ \Delta t $取值越小时, 补偿随机$ \theta $法稳定性效果会更好.

图 2 (a) 当$\theta = 0$时; (b)当$\theta = 0.2$时, 在不同步长$\Delta t = 1/20, 1/8, 1/2, 1$数值解的轨迹.
6 结论

本文主要讨论了带Poisson跳年龄相关随机时滞种群系统解析解和数值解的均方稳定性.给出系统解析解及数值解均方稳定的充分条件, 并展示出补偿随机$ \theta $法的均方稳定性.更精确地是, 当$ 1/2\leq\theta\leq1 $时, 对于任意的步长$ \Delta \tau/m $, 数值解是均方稳定的.当$ 0\leq\theta<1, $时, 如果步长$ \Delta t\in(0, \Delta t_0) $时, 数值解是指数均方稳定的.最后, 我们得出补偿随机$ \theta $法研究系统的稳定性可能不是最优的数值方法.以后我们将会采取不同的方法继续研究带跳年龄相关随机时滞种群系统的稳定性.

参考文献
[1] Whittle P, Arnold L. Stochastic differential equations:theory and applications[J]. J. Royal Stat. Soc., 1976, 139(3): 409. DOI:10.2307/2344856
[2] Jacod J, Shiryaev A N. Limit theorems for stochastic processes(2nd ed.)[M]. New York: Springer Verlag, 2002.
[3] Mao X R, Shah A. Exponential stability of stochastic differential equations[M]. New York: Marcel Dekker, 1994.
[4] Zhang Q M, Liu W A, Nie Z K. Existence, uniqueness and exponential stability for stochastic age-dependent population[J]. Appl. Math. Comput., 2004, 154(1): 183–201.
[5] Li R H, Leung P K, Pang W K. Convergence of numerical solutions to stochastic age-dependent population equations with Markovian switching[J]. Appl. Math. Comput., 2009, 233(4): 1046–1055. DOI:10.1016/j.cam.2009.08.113
[6] Ma W J, Zhang Q M, Han C Z. Numerical analysis for stochastic age-dependent population equations with fractional Brownian motion[J]. Commun. Nonl. Sci. Numer. Simul., 2012, 17(4): 1884–1893. DOI:10.1016/j.cnsns.2011.08.025
[7] Pang W K, Li R H, Liu M. Convergence of the semi-implicit Euler method for stochastic agedependent population equations[J]. Appl. Math. Comput., 2008, 195(2): 466–474.
[8] Sobczyk K. Stochastic differential equations with applications to physics and engineerin[M]. Netherlands: Springer, 1991.
[9] Li Q Y, Gan S Q. Stability of analytical and numerical solutions for nonlinear stochastic delay differential equations with jumps[J]. Abst. Appl. Anal., 2012, 2012(2): 183–194.
[10] Wang W Q, Chen Y P. Mean-square stability of semi-implicit Euler method for nonlinear neutral stochastic delay differential equations[J]. Appl. Numer. Math., 2011, 61(5): 696–701. DOI:10.1016/j.apnum.2011.01.003
[11] Wang L S, Wang X J. Convergence of the semi-implicit Euler method for stochastic age-dependent population equations with Poisson jumps[J]. Appl. Math. Model., 2010, 34(8): 2034–2043. DOI:10.1016/j.apm.2009.10.016
[12] 辛志贤, 张启敏, 李强, 哈金才. 具有年龄结构随机种群系统数值解的渐近有界性[J]. 数学杂志, 2018, 38(1): 147–154.
[13] 李强, 张启敏. 广义Khasminskii-type条件下与年龄相关随机时滞种群系统的数值解[J]. 数学的实践与认识, 2016, 46(18): 159–169.
[14] Ma W J, Zhang Q M, Wang Z P. Asymptotic stability of stochastic age-dependent population equations with Markovian switching[J]. Appl. Math. Comput., 2014, 227(227): 309–319.
[15] Ma W J, Ding B C, Zhang Q M. The existence and asymptotic behaviour of energy solutions to stochastic age-dependent population equations driven by Levy processes[J]. Appl. Math. Comput., 2015, 256: 656–665.
[16] Mao X R. tochastic differential equations and applications (2nd ed.) [M]. Chichester: Horwood Publishing, 2007.
[17] Baker C T H, Buckwar E. On Halanay-type analysis of exponential stability for the theta-Maruyama method for stochastic delay differential equations[J]. Stoch. Dynam., 2005, 5(2): 201–209. DOI:10.1142/S0219493705001341