数学杂志  2019, Vol. 39 Issue (4): 555-565   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
薛翔
王廷春
五次非线性Schrödinger方程的一个新型守恒紧致差分格式
薛翔, 王廷春    
南京信息工程大学数学与统计学院, 江苏 南京 210044
摘要:本文研究了带五次项的非线性Schrödinger方程初边值问题.利用有限差分法构造了一个四阶紧致差分格式,证明格式在离散意义下保持原问题的两个守恒性质,即质量守恒和能量守恒.引入“抬升”技巧,运用标准的能量方法和数学归纳法建立了误差的最优估计,证明数值解在空间和时间两个方向分别具有四阶和二阶精度.数值实验对理论结果进行了验证,并与已有结果进行了对比,结果表明本文格式在保持精度相当的前提下具有更高的计算效率.
关键词五次非线性Schrödinger方程    紧致有限差分格式    离散守恒律    最优误差估计    计算效率    
A NEW COMPACT FINITE DIFFERENCE SCHEME FOR THE QUANTIC NONLINEAR SCHRÖDINGER EQUATION
XUE Xiang, WANG Ting-chun    
School of Mathematics and Statistics, Nanjing University of Information Science and Technology, Nanjing 210044, China
Abstract: In this paper, we study the nonlinear Schrödinger equation with a quintic terma of the initial boundary value problem. By using the finite difference method to construct a fourth-order compact finite difference scheme, we prove that the scheme preserves the total mass and energy, respectively. By introducing the lifting technique, the optimal error estimate of the proposed scheme is established by using the standard energy method and the mathematical induction. It is proved that the numerical solution has accuracy of fourth-order and second-order in space and time, respectively. Numerical experiments are given to verify the theoretical results and compared with the existing results, which show that the proposed scheme has higher computational efficiency under the condition of maintaining high accuracy.
Keywords: quintic nonlinear Schrödinger equation     compact finite difference scheme     discrete conservation laws     optimal error estimate     computational efficiency    
1 引言

非线性Schrödinger方程在量子力学、非线性光学等物理领域有着非常广泛的应用.本文研究如下一类五次非线性Schrödinger方程的初边值问题[1]

$ \begin{align} &iu_t + u_{xx}-(|u|^{2}+|u|^{4}) u = f(x, t)u, \; \; \; \; (x, t)\in\; (a, b)\times(0, T], \end{align} $ (1.1)
$ \begin{align} &u(x, 0) = u_0(x), \; \; \; x\in\; [a, b], \end{align} $ (1.2)
$ \begin{align} &u(a, t) = u(b, t) = 0, \; \; \; t\in\; (0, T], \end{align} $ (1.3)

其中$ f(x, t) $为实值函数, $ u(x, t) $为复值函数, $ u_0 $为已知的复值函数.不难验证, 当$ f(x, t)\equiv f(x) $时, 该方程满足两个重要的守恒律, 即总质量和总能量守恒[1, 2]

$ \begin{eqnarray} &&Q(t) = \int_a^{b}\mid u \mid^{2}dx = Q(0), \end{eqnarray} $ (1.4)
$ \begin{eqnarray} &&E(t) = \int_a^{b}[|\frac{\partial u}{\partial x}|^{2}+f(x)|u|^2]dx+(G, 1) = E(0), \end{eqnarray} $ (1.5)

其中

$ (G, 1) = \int_a^{b}(\frac{1}{2}|u|^{4}+\frac{1}{3}|u|^{6})dx. $

关于非线性Schrödinger方程的数值研究已有很多结果. Bao等人[3-5]运用时间分裂拟谱方法对非线性Schrödinger/Gross-Pitaeviskill方程进行了数值求解, 并对某些物理现象进行了数值模拟; Argyris和Akrivis等人[6-9]运用有限元法对该方程进行了数值研究; Dehghan和Mirzaize运用无网格法对该方程进行了数值求解[10-11].文献[12-14]则运用谱方法和拟谱方法对该方程进行了数值求解和误差分析.因编程简单并能保持原问题的某些守恒性质, 有限差分法广泛应用于非线性Schrödinger方程的数值模拟[15-18].然而关于五次非线性Schrödinger方程的数值研究尚不多见.张鲁明、常谦顺等人[1-2]对该方程提出两个二阶有限差分格式, 证明其保持原问题的两个守恒性质, 同时建立了$ L^{2} $范数下的最优误差估计.王询等人[19]用待定系数的方法构建了一类五点有限差分格式, 该格式族在选取适当的参数后, 其计算精度在空间可达四阶, 然而计算中在每一个时间步都需要求解一个五对角代数方程组.为提高精度, 文献[20]提出一个紧致有限差分格式, 但该格式是非线性的, 计算中不可避免的需要迭代, 从而降低计算效率.另外, 该文作者也没有分析该格式是否在离散意义下保持原问题的两个守恒性质.鉴于以上分析, 本文旨在对问题(1.1)-(1.3)构造一个线性化紧致有限差分格式, 使得新格式在离散意义下依然保持原问题的两个守恒性质.

2 有限差分格式及其守恒律

对平面区域$ W = [a, b]\times[0, T] $进行网格剖分.取正整数$ J $, $ N $, 时空方向上的步长分别为$ \tau = \frac{T}{N}, h = \frac{b-a}{J} $, 网格点分别为

$ t_n = n\tau, x_j = a+jh\; (j = 0, 1, \cdots, J; \, n = 0, 1, \cdots, N), $

并记$ U_j^n = u(x_j, t_n), u_j^n\approx u(x_j, t_n) $.为书写简单, 引入以下记号

$ \begin{eqnarray*} &&\delta_x^{+}u_j^n = \frac{u_{j+1}^n-u_j^{n}}{h}, \delta_t^{+}u_j^n = \frac{u_{j}^{n+1}-u_j^{n}}{\tau}, \delta_x^{2}u_j^n = \frac{u_{j+1}^{n}-2u_j^{n}+u_{j-1}^{n}}{h^{2}}, \\ &&\mathcal{A}_h u_j^n = u_j^n + \frac{h^{2}}{12}\delta_x^{2}u_j^n = \frac{1}{12}(u_{j-1}^n+10u_j^n+u_{j+1}^n). \end{eqnarray*} $

定义$ X_h = \{w|w = (\omega_0, \omega_1, \cdots, \omega_{J-1}, \omega_J), \omega_0 = \omega_J = 0\} $为网格函数空间.设$ u, v $$ X_h $上的任意两个函数, 其内积和范数定义为

$ \langle u, v\rangle = h\sum\limits_{j = 1}^{J-1}u_{j}\bar{v_j}, \; \; \; \|u^{n}\| = (h\sum\limits_{j = 1}^{J-1}|u_j^n|^{2})^{\frac{1}{2}}. $

本文需要用到如下引理.

引理1 [21]  对于任意一个网格函数$ u\in X_h $, 有

$ \begin{equation} \|\delta_x^{+}u^{n}\|\leq\frac{4}{h}\|u^n\|, \|\delta_x^{2}u^n\|\leq\frac{4}{h}\|\delta_x^{+}u^n\|. \end{equation} $ (2.1)

引理2 [22]  如果网格函数$ u\in X_h $, 则

$ \begin{equation} C_1\|u^n\|\leq\|\mathcal{A}_h^{-1}u^n\|\leq C_2\|u^n\|. \end{equation} $ (2.2)

引理3 [21]  任给网格函数$ u\in\; X_h $, 有

$ \begin{equation} \|u^n\|_\infty\leq\frac{\sqrt{b-a}}{2}\|\delta_x^{+}u^n\|. \end{equation} $ (2.3)
2.1 有限差分格式

对初边值问题(1.1)-(1.3), 本文提出如下四阶紧致差分格式

$ \begin{align} &i\mathcal{A}_h\delta_t^{+}u_j^n+\frac{1}{2}\delta_x^2(u_j^{n+1}+u_j^n)-\frac{1}{2}\mathcal{A}_h[(\frac{3}{2}(|u_j^n|^2+|u_j^n|^4) -\frac{1}{2}(|u_j^{n-1}|^2+|u_j^{n-1}|^4)) \end{align} $ (2.4)
$ \begin{align} &\times(u_j^{n+1}+u_j^n)] = \frac{1}{2}\mathcal{A}_h[f_j^{n+\frac{1}{2}}(u_j^{n+1}+u_j^{n})], \; \; j = 1, 2, \cdots, J-1; n = 1, 2, \cdots, N-1, \\ &u_j^0 = u_0(x_j), \; \; \; j = 0, 1, 2, \cdots, J, \end{align} $ (2.5)
$ \begin{align} &u_0^n = u_J^n = 0, \; \; \; n = 1, 2, \cdots, N, \end{align} $ (2.6)

其中$ f_j^{n+\frac{1}{2}} = f(x_j, t_{n+\frac{1}{2}}), j = 1, 2, \cdots, J-1; \, n = 1, 2, \cdots, N-1 $.显见$ (2.4) $式是一个三层的格式, 无法自启动, 为求解第一层数值解, 给出如下方法

$ \begin{equation} u_j^1 = u_0(x_j)+\tau\tilde{u}(x_j), \; \; \; j = 1, 2, \cdots, J-1, \end{equation} $ (2.7)

其中

$ \begin{align} &\tilde{u_j} = i[\partial_{xx}u_0(x_j)-(|u_0(x_j)|^2+|u_0(x_j)|^4)u_0(x_j)-f(x_j, 0)u_0(x_j)], \\ &j = 1, 2, \cdots, J-1. \end{align} $ (2.8)

在(2.4)式两端同时作用算子$ \mathcal{A}_h^{-1} $, 则(2.4)式可写成如下等价形式

$ \begin{equation} \begin{split} &i\delta_t^{+}u_j^n+\frac{1}{2}\mathcal{A}_h^{-1}\delta_x^2(u_j^{n+1}+u_j^n)-\frac{1}{2}[(\frac{3}{2}(|u_j^n|^2+|u_j^n|^4) -\frac{1}{2}(|u_j^{n-1}|^2+|u_j^{n-1}|^4))\\ &\times(u_j^{n+1}+u_j^n)] = \frac{1}{2}[f_j^{n+\frac{1}{2}}(u_j^{n+1}+u_j^{n})], \; \; j = 1, 2, \cdots, J-1; \, n = 1, 2, \cdots, N-1. \end{split} \end{equation} $ (2.9)
2.2 离散守恒律

为证明离散守恒律, 定义

$ \begin{equation} {\langle F^n, 1\rangle : = h\sum\limits_{j = 1}^{J-1}F_j^n, \; \; \; n = 0, 1, \cdots, N, } \end{equation} $ (2.10)

其中

$ \begin{align} &{F_j^{n+1}: = F_j^{n}+[\frac{3}{2}(|u_j^n|^2+|u_j^n|^4)-\frac{1}{2}(|u_j^{n-1}|^2+u_j^{n-1}|^4)](|u_j^{n+1}|^2-|u_j^{n}|^2), } \end{align} $ (2.11)
$ \begin{align} &n = 1, 2, \cdots, N-1, \\ &{F_j^l: = (\frac{1}{2}|u_j^0|^4+\frac{1}{3}|u_j^0|^6), \; \; \; j = 1, 2, \cdots, J-1;\, \; \; \; l = 0, 1.} \end{align} $ (2.12)

引理4 (Ⅰ)   格式(2.4)-(2.6)在离散意义下满足总质量守恒, 即

$ \begin{equation} Q^n = \|u^n\|^2 = Q^1, \; \; n = 1, 2, \cdots, N. \end{equation} $ (2.13)

(Ⅱ)如果$ f $不依赖于时间(即有$ f_j^{k-\frac{1}{2}}\equiv f_j $), 则格式(2.4)-(2.6)在离散意义下满足总能量守恒, 即

$ \begin{equation} {E^n = \||\delta_x^{+}u^n\||^2+h\sum\limits_{j = 1}^{J-1}f_j|u_j^n|^2+\langle F^n, 1\rangle = E^1, \; \; n = 1, 2, \cdots, N, } \end{equation} $ (2.14)

其中

$ \||\delta_x^{+}u^n\||: = -\langle \mathcal{A}_h^{-1}\delta_x^{2}u^n, u^n\rangle^\frac{1}{2}. $

  将$ (2.9) $式与$ u^{n+1}+u^{n} $做内积, 并取虚部得

$ \begin{equation} \frac{1}{\tau}(\|u^{n+1}\|^2-\|u^n\|^2) = 0, n = 1, 2, \cdots, N-1. \end{equation} $ (2.15)

$ Q^n: = \|u^n\|^2 $, 则有

$ \begin{equation} Q^{n+1} = Q^n, n = 1, 2, \cdots, N-1. \end{equation} $ (2.16)

$ \begin{equation} Q^n = Q^1, n = 1, 2, \cdots, N. \end{equation} $ (2.17)

将(2.9)式与$ u^{n+1}-u^n $做内积, 并取实部得

$ \begin{equation} \begin{split} &{\rm Re}\langle i\delta_t^{+}u^n, u^{n+1}-u^n\rangle+\frac{1}{2}{\rm Re}\langle\mathcal{A}_h^{-1}\delta_x^2(u^{n+1}+u^n), u^{n+1}-u^n\rangle\\ &-\frac{1}{4}{\rm Re}\langle[3(|u^n|^2+|u^n|^4)-(|u^{n-1}|^2+|u^{n-1}|^4)](u^{n+1}+u^n), u^{n+1}-u^n\rangle\\ = &\frac{1}{2}{\rm Re}\langle f(u^{n+1}+u^n), u^{n+1}-u^n\rangle, \; \; n = 1, 2, \cdots, N-1. \end{split} \end{equation} $ (2.18)

直接计算可得

$ \begin{eqnarray} &&{\rm Re}\langle i\delta_t^{+}u^n, u^{n+1}-u^n\rangle = {\rm Re}\{\frac{i}{\tau}\langle u^{n+1}-u^n, u^{n+1}-u^{n}\rangle\} = 0, \\ &&{n = 1, 2, \cdots, N-1, } \end{eqnarray} $ (2.19)
$ \begin{eqnarray} &&\frac{1}{2}{\rm Re}\langle \mathcal{A}_h^{-1}\delta_x^2(u^{n+1}+u^n), u^{n+1}-u^n\rangle = -\frac{1}{2}(\||\delta_x^{+}u^{n+1}\||^2-\||\delta_x^{+}u^{n}\||^2), \\ &&{n = 1, 2, \cdots, N-1, } \end{eqnarray} $ (2.20)
$ \begin{eqnarray} &&-\frac{1}{4}{\rm Re}\langle[3(|u^n|^{2}+|u^n|^{4})-(|u^{n-1}|^{2}+|u^{n-1}|^{4})](u^{n+1}+u^{n}), u^{n+1}-u^{n}\rangle\\ && = -\frac{1}{4}\sum\limits_{j = 1}^{J-1}h[3|u_j^n|^{2}+|u_j^n|^{4})-(|u_j^{n-1}|^2+|u_j^{n-1}|^{4})](|u_j^{n+1}|^2-|u_j^n|^2), \end{eqnarray} $ (2.21)
$ \begin{array}{l} {n = 1, 2, \cdots, N-1, } \\ \frac{1}{2}{\rm Re}\langle f(u^{n+1}+u^n), u^{n+1}-u^n\rangle = \frac{h}{2}\sum\limits_{j = 1}^{J-1}f_j(|u_j^{n+1}|^{2}-|u_j^n|^{2}), \\ n = 1, 2, \cdots, N-1, \end{array} $ (2.22)

将(2.19)-(2.22)式代入$ (2.18) $式得

$ \begin{equation} \begin{split} &\||\delta_x^{+}u^{n+1}\||^2-\||\delta_x^{+}u^{n}\||^2+h\sum\limits_{j = 1}^{J-1}f_j(|u_j^{n+1}|^{2}-|u_j^n|^{2}) +\langle F^{n+1}-F^{n}, 1\rangle = 0, \\ &{n = 1, 2, \cdots, N-1. } \end{split} \end{equation} $ (2.23)

$ \begin{equation} E^n: = \||\delta_x^{+}u^{n}\||^2+h\sum\limits_{j = 1}^{J-1}f_j|u_j^{n}|^{2}+\langle F^n, 1\rangle, n = 1, \cdots, N-1. \end{equation} $ (2.24)

则得

$ \begin{equation} E^{n+1}-E^{n} = 0, \; \; \; {n = 1, 2, \cdots, N-1.} \end{equation} $ (2.25)

$ \begin{equation} E^{n} = E^{1}, \; \; \; {n = 1, 2, \cdots, N.} \end{equation} $ (2.26)
3 误差估计

记格式(2.4)-(2.7)的局部截断误差为

$ R_j^n\; (j = 1, 2, \cdots, J-1; \, n = 0, 1, \cdots, N-1), $

其定义如下

$ \begin{align} R_j^{0} = &u(x_j, \tau)-u_0(x_j)-\tau\tilde{u_j}, j = 1, 2, \cdots, J-1, \end{align} $ (3.1)
$ \begin{align} R_j^{n} = &i\delta_t^{+}U_j^n+\frac{1}{2}\mathcal{A}_h^{-1}\delta_x^2(U_j^{n+1}+U_j^n)-\frac{1}{2}[(\frac{3}{2}(|U_j^n|^2+|U_j^n|^4) \\ &-\frac{1}{2}(|U_j^{n-1}|^2+|U_j^{n-1}|^4))(U_j^{n+1}+U_j^n)]-\frac{1}{2}[f_j^{n+\frac{1}{2}}(U_j^{n+1}+U_j^{n})], \\ \nonumber &j = 1, 2, \cdots, J-1; \, n = 1, 2, \cdots, N-1, \end{align} $ (3.2)

其中$ U_j^{n} = u(x_j, t_n) $.运用Taylor展开可得如下引理.

引理5 格式(2.4)-(2.7)的局部截断误差满足

$ \begin{align} &R_j^{0} = O(\tau^{2}), \; \; \; j = 1, 2, \cdots, J-1, \end{align} $ (3.3)
$ \begin{align} &R_j^{n} = O(\tau^{2}+h^4), \; \; \; j = 1, 2, \cdots, J-1; \, n = 1, 2, \cdots, N-1. \end{align} $ (3.4)

定义误差函数$ e_j^n = U_j^{n}-u_j^n, \; \; j = 0, 1, 2, \cdots, J; \, n = 0, 1, 2, \cdots, N, $则有

定理1   假设$ u\in C^{6, 3}([a, b]\times[0, T]) $, 则差分格式(2.4)-(2.7)的解以$ \|\cdot\| $范数收敛到初边值问题(1.1)-(1.3)的解, 收敛阶为$ O(\tau^{2}+h^{4}) $.

  将$ (3.1) $式和$ (3.2) $式分别与$ (2.7) $式和$ (2.9) $式相减可得如下误差方程

$ \begin{align} &e_j^1 = R_j^0, j = 1, 2, \cdots, J-1, \end{align} $ (3.5)
$ \begin{align} &i\delta_t^{+}e_j^n+\frac{1}{2}\mathcal{A}_h^{-1}\delta_x^2(e_j^{n+1}+e_j^n)-\frac{1}{4}\chi_j^{n+1}-\frac{1}{2}[f_j^{n+\frac{1}{2}}(e_j^{n+1}+e_j^{n})] = R_j^n, \\ \nonumber &j = 1, 2, \cdots, J-1; \, n = 1, 2, \cdots, N-1, \end{align} $ (3.6)

其中

$ \begin{eqnarray} \chi_j^{n+1}& = &(3|U_j^n|^{2}+3|U_j^n|^{4}-|U_j^{n-1}|^{2}-|U_j^{n-1}|^4)(U_j^n+U_j^{n+1})-(3|u_j^n|^2+3|u_j^n|^{4}\\ &&-|u_j^{n-1}|^{2}-|u_j^{n-1}|^4)(u_j^{n}+u_j^{n+1})\\ & = &(3|U_j^n|^{2}+3|U_j^n|^{4}-|U_j^{n-1}|^{2}-|U_j^{n-1}|^4)(e_j^n+e_j^{n+1})+[3U_j^{n}\bar{e}_j^n+3e_j^n\bar{u}_j^n\\ &&+3(|U_j^n|^{2}+|u_j^n|^{2})(U_j^{n}\bar{e}_j^n+e_j^n\bar{u}_j^n)-U_j^{n-1}\bar{e}_j^{n-1}-e_j^{n-1}\bar{u}_j^{n-1}-(|U_j^{n-1}|^{2}+|u_j^{n-1}|^{2})\\ &&\times(U_j^{n-1}\bar{e}_j^{n-1}+e_j^{n-1}\bar{u}_j^{n-1})](u_j^n+u_j^{n+1}).\\ &&j = 1, 2, \cdots, J-1; \, n = 1, 2, \cdots, N-1. \end{eqnarray} $ (3.7)

$ (3.5) $式和引理$ 5 $显见

$ \begin{equation} \|e^1\|\leq C\tau^2. \end{equation} $ (23.8)

运用Taylor展开, 可得

$ \begin{equation} \delta_x^{+}R_j^0\leq O(\tau^2), \end{equation} $ (3.9)
$ \begin{equation} \delta_x^{2}R_j^0\leq O(\tau^2). \end{equation} $ (3.10)

因此有

$ \begin{eqnarray} &&\|\delta_x^{+}e^1\| = \|\delta_x^{+}R^0\|\leq C\tau^2, \end{eqnarray} $ (3.11)
$ \begin{eqnarray} && \|\delta_x^{2}e^1\| = \|\delta_x^{2}R^0\|\leq C\tau^2, \end{eqnarray} $ (3.12)
$ \begin{eqnarray} && \|e^1\|_\infty\leq\|e^1\|+\|\delta_x^{+}e^1\|\leq C\tau^2\leq C(\tau^2+h^4). \end{eqnarray} $ (3.13)

现假设当$ n\leq k\; (1\leq k<N) $时, 有

$ \begin{equation} \|e^n\|\leq C(\tau^2+h^4), \; \; \|\delta_x^{2}e^n\|\leq C. \end{equation} $ (3.14)

进而可得

$ \begin{eqnarray} &&\|u^n\|_\infty\leq\|U^n\|_\infty+\|e^n\|_\infty\leq C, \end{eqnarray} $ (3.15)
$ \begin{eqnarray} && \|\chi^{n+1}\|\leq C(\|e^{n-1}\|+\|e^n\|+\|e^{n+1}\|)\leq C(\tau^2+h^4). \end{eqnarray} $ (3.16)

$ (3.6) $式与$ e^n+e^{n+1} $做内积, 取虚部得

$ \begin{equation} \begin{split} &\frac{1}{\tau}(\|e^{n+1}\|^2-\|e^n\|^2)+\frac{1}{2}{\rm Im}\langle\mathcal{A}_h^{-1}\delta_x^{2}(e^n+e^{n+1}), e^n+e^{n+1}\rangle-\frac{1}{4}{\rm Im}\langle\chi^{n+1}, e^{n}+e^{n+1}\rangle\\ &-\frac{1}{2}{\rm Im}\langle f^{n+\frac{1}{2}}(e^{n}+e^{n+1}), e^{n}+e^{n+1}\rangle = {\rm Im}\langle R^n, e^n+e^{n+1}\rangle. \end{split} \end{equation} $ (3.17)

$ \begin{eqnarray} &&{\rm Im}\langle\mathcal{A}_h^{-1}\delta_x^{2}(e^n+e^{n+1}), e^n+e^{n+1}\rangle = 0, \end{eqnarray} $ (3.18)
$ \begin{eqnarray} &&{\rm Im}\langle f^{n+\frac{1}{2}}(e^{n}+e^{n+1}), e^{n}+e^{n+1}\rangle = 0, \end{eqnarray} $ (3.19)
$ \begin{eqnarray} &&{\rm Im}\langle R^n, e^n+e^{n+1}\rangle\leq C(\|R^n\|+\|e^n\|+\|e^{n+1}\|), \end{eqnarray} $ (3.20)
$ \begin{eqnarray} && |\langle \chi^{n+1}, e^n+e^{n+1}\rangle|\leq C(\|e^{n-1}\|^{2}+\|e^n\|^{2}+\|e^{n+1}\|^2), \end{eqnarray} $ (3.21)

由(3.18)-(3.21)式, 得

$ \begin{equation} \|e^{n+1}\|^{2}-\|e^n\|^{2}\leq C\tau(\|e^{n-1}\|^{2}+\|e^n\|^{2}+\|e^{n+1}\|^2)+C\tau\|R^n\|^2. \end{equation} $ (3.22)

上式对$ n $求和, 有

$ \begin{equation} \|e^{n+1}\|^{2}-\|e^1\|^{2}\leq C\tau\sum\limits_{l = 1}^{n+1}\|e^l\|^{2}+C(\tau^2+h^4)^2. \end{equation} $ (3.23)

$ \tau $足够小时, 由Gronwall不等式得

$ \begin{equation} \|e^{n+1}\|\leq C(\tau^2+h^4). \end{equation} $ (3.24)

$ (3.6) $式可得

$ \begin{equation} \begin{split} &\frac{1}{2}\|\mathcal{A}_h^{-1}\delta_x^{2}(e^{n}+e^{n+1})\|\\ = &\|-i\delta_t^{+}e^n+\frac{1}{4}\chi^{n+1}+\frac{1}{2}[f^{n+\frac{1}{2}}(e^{n}+e^{n+1})]+R^n\|\\ \leq& C\|\delta_t^{+}e^{n}\|+C\|\chi^{n+1}\|+C\|f^{n+\frac{1}{2}}(e^{n}+e^{n+1})\|+\|R^n\|\\ \leq& C\tau^{-1}(\|e^{n+1}\|+\|e^n\|)+C(\tau^2+h^4)\\ \leq& C\tau^{-1}(\tau^2+h^4). \end{split} \end{equation} $ (3.25)

运用引理$ 2 $

$ \begin{equation} \begin{split} \|\delta_x^{2}e^{n+1}\|-\|\delta_x^{2}e^n\|\leq\|\delta_x^{2}(e^{n+1}+e^n)\| \leq C\|\mathcal{A}_h^{-1}\delta_x^{2}(e^{n+1}+e^n)\|\leq C\tau^{-1}(\tau^2+h^4), \end{split} \end{equation} $ (3.26)

上式对$ n $求和得

$ \begin{equation} \|\delta_x^{2}e^{n+1}\|\leq C(n+1)\tau^{-1}(\tau^2+h^4)\leq C\tau^{-2}(\tau^2+h^4)\leq C+C\tau^{-2}h^4. \end{equation} $ (3.27)

只要当$ h\leq\tau $时, 有$ h^4/\tau^2\leq1 $, 此时有

$ \begin{equation} \|\delta_x^{2}e^{n+1}\|\leq C. \end{equation} $ (3.28)

另外由引理$ 1 $可得

$ \begin{equation} \|\delta_x^{2}e^{n+1}\|\leq\frac{16}{h^2}\|e^{n+1}\|\leq Ch^{-2}(h^4+\tau^2)\leq C(h^2+\tau^{2}h^{-2}), \end{equation} $ (3.29)

由此可见, 当$ \tau\leq h $时, 有$ \tau^2/h^2\leq 1 $, 得

$ \begin{equation} \|\delta_x^{2}e^{n+1}\|\leq C. \end{equation} $ (3.30)

所以无论网格比如何, 总有

$ \|\delta_x^{+}e^{n+1}\|\leq C\|\delta_x^{2}e^{n+1}\|\leq C. $

$ (3.24) $式和$ (3.30) $式可知假设$ (3.14) $式对$ n = k+1 $时也成立.证毕.

4 数值实验

便于验证格式的精度, 引入以下记号

$ \begin{eqnarray*} &&E(h, \tau) = \|e(h, \tau)\|, \; \; \text {rate}1 = \log_2\frac{E(h, \tau)}{E(h/2, \tau)}, \; \; \text {rate}2 = \log_2\frac{E(h, \tau)}{E(h, \tau/2)}, \\ &&\text {order}1 = \frac{\log(\|E(h_1, \tau)\|/\|E(h_2, \tau)\|)}{\log(h_1/h_2)}, \, \text {order}2 = \frac{\log(\|E(h, \tau_1)\|/\|E(h, \tau_2)\|)}{\log(\tau_1/\tau_2)}. \end{eqnarray*} $

算例4.1   考虑如下初边值问题

$ \begin{equation} \begin{cases} iu_{t}+u_{xx}-(|u|^{2}+|u|^{4})u = f(x, t)u, \; \; {(x, t)\in(a, b)\times(0, 1], \nonumber} \\ u(x, 0) = u_0(x) = \exp(-x^{2}+ix), \; \; {x\in[a, b], \nonumber} \end{cases} \label{bbmb1} \end{equation} $

其中

$ f(x, t) = 4(x-2t)^{2}-\exp[-2(x-2t)^2]-\exp[-4(x-2t)^2]. $

该问题的精确解为

$ u(x, t) = \exp(-(x-2t)^2+i(x-3t)). $

由上式可见当$ x $趋向于无穷大时, $ u(x, t) $迅速趋于0.故当$ -a $$ b $取得足够大时, $ u(a, t) $$ u(b, t) $近似为0.因此在数值求解时取空间方向的计算区间为$ (-15, 15) $, 以忽略截断误差的影响.

在验证空间方向(时间方向)收敛阶时, 取$ \tau = 0.00001\; (h = 0.0001) $, 这样可以忽略时间(空间)方向的误差影响.表 1表 2分别给出了空间和时间方向的收敛阶.在表 3中, 将本文的线性格式与文献[20]中的非线性格式做了计算效率上的比较.图 1展示了精确解和数值解在不同时间层下的波形变化.

表 1 空间方向收敛阶

表 2 时间方向收敛阶

表 3 本文中的线性格式(LCFD)与文[20]中的非线性格式(NCFD)在计算效率上的比较

图 1 算例4.1在$ \tau = 0.01, h = 0.1 $时的精确解(左)和数值解(右)

算例4.2   对形如$ iu_t+u_{xx}+\sigma(x, t)u+(\beta_1|u|^2+\beta_2|u|^4)u = f(x, t), \; (x, t)\in(0, \pi)\times(0, 1]. $$ \beta_1 = \beta_2 = -1, \sigma(x, t) = \frac{1}{2}x^2, f(x, t) = [-\frac{1}{2}x^2+(|e^{it}\sin x|^2+|e^{it}\sin x|^4)-2]e^{it}\sin x, $初始值为$ u_0(x) = \sin x $, 则精确解为$ u(x, t) = e^{it}\sin x $.

运用本文中的格式(2.4)-(2.6)对算例$ 4.2 $进行求解, 有以下结果

表 4 空间方向取不同步长时的误差和精度

表 5 时间方向取不同步长时的误差和精度

由以上数值实验的结果可看出:差分格式(2.4)-(2.6)在时空方向分别具有$ 2 $阶和$ 4 $阶精度, 而且格式在离散意义下依然能够保持总质量和总能量守恒, 这完全符合定理$ 1 $和引理$ 4 $的结论.除此之外, 与已有格式相比, 本文格式在精度相当的前提下还大幅提高了计算效率.

图 2 算例4.2在不同时间层的总质量(左)和总能量(右)

图 3 算例4.2在离散意义下的总质量守恒(左)和总能量守恒(右)
参考文献
[1] 张鲁明, 常谦顺. 带五次项的非线性Schrödinger方程的守恒数值格式[J]. 应用数学, 1999, 12: 65–71.
[2] 张鲁明, 常谦顺. 带五次项的非线性Schrödinger方程的差分解法[J]. 应用数学学报, 2000, 23: 351–358.
[3] Bao W, Jaksch D, Markowich P A. Numerical solution of the Gross-Pitaevskii equation for BoseEinstein condensation[J]. J. Comput. Phys., 2003, 187: 318–342. DOI:10.1016/S0021-9991(03)00102-5
[4] Bao W, Li H, Shen J. A generalized-Lagueree-Fourier-Hermite pseudospectral method for computing the dynamics of rotating Bose-Einstein condensates[J]. SIAM. J. Sci. Comput., 2009, 31: 3685–3711. DOI:10.1137/080739811
[5] Bao W, Shen J. A Fourth-order time-splitting Lagueree-Hermite psectralmethod for Bose-Einstein condensates[J]. SIAM J. Sci. Comput., 2005, 26: 2010–2028.
[6] Argyris J, Haase M. An engineer's guide to solution phenomena:application of the finite element method[J]. Comput. Meth. Appl. Mech. Engrg., 1987, 61: 71–122. DOI:10.1016/0045-7825(87)90117-4
[7] Akrivis G, Dougalis V, Karakashian O. On fully discrete Galerkin methods for second order temporal accuracy for the nonlinear Schrödinger equation[J]. Numer. Math., 1991, 59: 31–53. DOI:10.1007/BF01385769
[8] Karakashian O, Makridakis C. A space-time finite element method for the nonlinear Schrödinger equation:the discontinuous Galerkin method[J]. Math. Comp., 1998, 67: 479–499. DOI:10.1090/S0025-5718-98-00946-6
[9] Xu Y, Shu C W. Local discontinuous Galerkin methods for nonlinear Schrödinger equation[J]. J. Comput. Phys., 2005, 205: 72–77. DOI:10.1016/j.jcp.2004.11.001
[10] Dehghan M, Mirzaei D. Numerical solution to the unsteady two-dimensional Schrödinger equation using meshless local boundary integral equation method[J]. Intern. J. Numer. Meth. Engin., 2008, 76: 501–520. DOI:10.1002/nme.2338
[11] Dehghan M, Mirzaei D. The meshless local Petrov-Galerkin (MLPG) method for the generalizad two-dimensional non-linear Schrödinger equation[J]. Engin. Anal. Bound. Elem., 2008, 32: 747–756. DOI:10.1016/j.enganabound.2007.11.005
[12] Taha T R, Ablowitz M J. Analytical and numerical aspects of certain nonlinear evolution equations[J]. J. Comput. Phys., 1984, 5: 203–230.
[13] Cloot A, Herbet B M, Weideman J A C. A numerical study of the nonlinear Schrödinger equation involving quintic terms[J]. J. Comput. Phys., 1990, 86: 127–146. DOI:10.1016/0021-9991(90)90094-H
[14] Dehghan M, Taleei A. A compact split-step finite difference method for solving the nonlinear Schrödinger equation with constant and variable coefficients[J]. Comput. Phys. Comm., 2010, 181: 43–51. DOI:10.1016/j.cpc.2009.08.015
[15] 王廷春, 郭柏灵. 一维非线性Schrödinger方程的两个无条件收敛的守恒紧致差分格式[J]. 中国科学:数学, 2011, 41: 207–233.
[16] Sanz-Sema J M. Methods for the numerical solution of the nonlinear Schrödinger equation[J]. Math. Comp., 1984, 43: 21–27. DOI:10.1090/S0025-5718-1984-0744922-X
[17] Chang Q, Jia E, Sun W. Difference schemes for solving the generalized nonlinear Schrödinger equation[J]. J. Comput. Phys., 1999, 148: 397–415. DOI:10.1006/jcph.1998.6120
[18] Delfour M, Fortin M, Payre G. Finite-difference Solution of a Non-linear Schrödinger Equattion[J]. J. Comput. Phys., 1981, 44: 277–288. DOI:10.1016/0021-9991(81)90052-8
[19] 王询, 曹圣山. 带五次项的非线性Schrödinger方程新差分格式[J]. 中国海洋大学学报, 2009, 39: 487–491.
[20] 初日辉. 带五次项的非线性Schrödinger方程的一个紧致差分格式[J]. 江苏师范大学学报, 2014, 32: 53–57.
[21] Samarskii A A, Andreev V B. Difference methods for elliptic equations[M]. Moscow: Nauka, 1976.
[22] Wang T, Zhao X. Optimal l error estimates of finite difference methods for the coupled GrossPitaevskii equations in high dimensions[J]. Sci. China Math., 2014, 57(10): 2189–2214. DOI:10.1007/s11425-014-4773-7