数学物理学报, 2023, 43(3): 883-895

一类椭圆型界面问题的数值算法

于一康, 牛晶,*

哈尔滨师范大学数学科学学院 哈尔滨 150025

A Kind of Numerical Algorithm for Elliptic Interface Problem

Yu Yikang, Niu Jing,*

School of Mathematics and Sciences, Harbin Normal University, Harbin 150025

通讯作者: *牛晶,E-mail: qq63192678@126.com

收稿日期: 2022-11-16   修回日期: 2023-01-12  

基金资助: 国家青年自然科学基金项目(12101164)
哈尔滨师范大学硕士研究生创新科研项目(HSDSSCX2022-40)

Received: 2022-11-16   Revised: 2023-01-12  

Fund supported: Youth Fund of NSFC(12101164)
Postgraduate Innovative Scientific Research Project of Harbin Normal University(HSDSSCX2022-40)

摘要

该文基于再生核空间相关理论提出了一种一维椭圆界面问题的数值算法. 该方法通过积分${W}^{1}_{2}$空间的再生核函数, 得到了三次样条空间的一组新基, 在此基础上建立了一个破裂的三次样条空间, 然后利用最小二乘法得出了此类界面问题的近似解, 并且讨论了在$H_2$范数、$H_1$范数和$L_2$范数下的收敛阶, 最后通过几个数值算例得到了验证.

关键词: 界面问题; 再生核函数; 三次样条空间; 最小二乘法; 收敛性分析

Abstract

In this paper, a numerical algorithm based on the theory of reproducing kernel space is proposed for the one-dimensional ellipse interface problem. This method integrates the reproducing kernel function of ${W}^{1}_{2}$ space to obtain a set of new basis in the cubic spline space, on which a broken cubic spline space is established. Then we use the least squares method to get the approximate solution of this kind of interface problem. We discussed the order of convergence under the $H_2$ norm, $H_1$ norm and $L_2$ norm. Finally, our theory is verified by several numerical example.

Keywords: Interface problem; Reproducing kernel; The cubic spline space; The least squares method; Convergence analysis

PDF (464KB) 元数据 多维度评价 相关文章 导出 EndNote| Ris| Bibtex  收藏本文

本文引用格式

于一康, 牛晶. 一类椭圆型界面问题的数值算法[J]. 数学物理学报, 2023, 43(3): 883-895

Yu Yikang, Niu Jing. A Kind of Numerical Algorithm for Elliptic Interface Problem[J]. Acta Mathematica Scientia, 2023, 43(3): 883-895

1 引言

椭圆型界面问题由于在生物化学[1]、工程和材料科学[2]中有着广泛的应用而备受关注. 关于椭圆型界面问题的理论研究, 如解的存在唯一性以及解的极限行为在文献[3-6]中均有研究.

在椭圆型界面问题中, 通常由不同的材料构成模拟区域, 这些区域由曲线或曲面隔开. 在许多文献中已经讨论了处理跳跃条件的数值方法, 例如有限差分法[5,7-8], 有限元方法[9-12]. 有限元方法虽然经常被使用, 但通常的平均误差估计不能保证在界面附近或在界面处的解的精度. 对于有限差分方法, 混合导数的离散化和收敛分析是一个具有挑战性的问题. 值得一提的是, 董、冯等人在文献[13] 中提出了一种新的有限元-有限差分(FE-FD) 方法,该方法将系数矩阵为对称半正定的有限元离散(远离界面)和系数矩阵部分具有M 矩阵性质的有限差分(靠近界面或在界面上)相结合, 采用基于浸没界面法的插值法, 精确地计算了界面两侧解(或梯度)的法向导数. 此外, 潘、吴等人在文献[14]中提出了界面拟合网格生成(EIFMG)算法, 和保线性有限体积(FV)格式, EIFMG算法通过弯曲单元边缘以与材料界面对齐, 在不改变单元总数和相关拓扑的情况下生成与材料界面精确匹配的结构化曲面四边形网格. 其次, 通过使用所谓的线性保持准则, 导出的FV格式即使在极粗的网格上也能在具有线性解的弯曲四边形网格上精确求解, 这通常是有限元或有限差分方法不容易实现的. 在文献[15-16] 中, 作者通过改变界面所在单元的基函数设计了浸没有限元方法(IFEM). 同样的方法也应用于浸没有限体积法[17-18].

而对于再生核方法(RKM)[19-22], 它可以获得更大步长的高精度光滑数值解, 并且对边界条件的处理也比较简单. 关于再生核方法, 牛、徐、于等人也做了很多研究[23-28]. 本文就是基于再生核空间相关理论利用最小二乘法求得如下界面问题的近似解, 得到了较高的计算精度.

假定存在界面$\Omega\in(0,1)$, 函数$\beta_1, \beta_2$, $\gamma_1, \gamma_2$足够光滑,有

$\begin{equation}\label{eq:1} \left\{\begin{array}{ll} (\beta_{1}f')'-\gamma_{1}f=h_{1}(x),& x\in\Omega_1,\\ (\beta_{2}f')'-\gamma_{2}f=h_{2}(x),& x\in\Omega_2. \end{array}\right. \end{equation}$

其中$\Omega_1=(0,\alpha)$, $\Omega_2=(\alpha,1)$, $h_{1}$$h_{2}$分别是$\Omega_1$$\Omega_2$上的一阶连续函数. 满足狄利克雷边界条件

$\begin{equation}\label{eq:2} f(0)=A,\qquad f(1)=B \end{equation}$

及界面$\Omega$上的界面条件

$\begin{equation}\label{eq:3} \left\{\begin{array}{ll} {}[f]|_{\Omega}=f(\alpha^{+})-f(\alpha^{-})=\delta_1,\\ {}[\beta f']|_{\Omega}=\beta_{2}(\alpha^{+})f'(\alpha^+)-\beta_{1}(\alpha^{-})f'(\alpha^{-})=\delta_2. \end{array}\right. \end{equation}$

其中, $[f]|_{\Omega}=f(\alpha^{+})-f(\alpha^{-})$等表示$f$在界点$\alpha$处满足界面$\Omega$的跳跃条件.

由于解的导数的不连续性, 传统的再生核空间${W}^{m}_{2}[0,1]$对界面问题(1.1)-(1.3)不适用. 本文则以一维椭圆型界面问题为例, 利用再生核的优良特性构造一个合适的空间-破裂三次样条空间, 并将最小二乘法和样条插值应用到此问题中, 以此期望得到更高精度的数值解方法, 从而用此方法求解其他界面问题. 全文共6节, 第1节介绍了界面问题的研究背景以及一些传统的数值算法, 并对本论文所研究的问题模型进行说明. 第2节给出了论文中所用到的再生核空间和基本知识, 并且构造出了一个破裂的三次样条空间. 第3 节提出了最小二乘法并且对该方法的可行性进行了证明. 第4节提出了范数误差估计, 证明了算法的收敛性. 第5节用数值算例支持了我们的理论发现. 第6节对全文进行总结.

2 预备知识

${\bf定义2.1}$[29]$H$$Hilbert$空间, $H$中的元素是集合$X$中的复值函数, 若存在唯一的函数 $K_x(y)\in H$, 对$\forall x\in X$满足 $ \left\langle f, K_{x}\right\rangle=f(x), \quad f \in H $ 则称$H$为再生核空间, 称$K(x,y)=K_x(y)$为再生核函数.

${\bf定义2.2}$ [29] ${W}^1_{2}[a,b]=\left\{f(x)|f(x)\mbox{在$[a,b]$上是绝对连续的实值函数}, f'(x) \in L^{2}[a,b]\right\}$ 是具有再生核$K(x,y)$的再生核空间.其内积和范数表示为

$\left\{\begin{array}{l} \langle f, g\rangle_{W_{2}^{1}[a, b]}=f(a) g(a)+\int_{a}^{b} f^{\prime} g^{\prime} \mathrm{d} x, f, g \in W_{2}^{1}[a, b], \\ \|f\|_{W_{2}^{1}[a, b]}^{2}=\langle f, f\rangle_{W_{2}^{1}[a, b]}, \quad \forall f \in H. \end{array}\right.$

${W}^{1}_{2}[a,b]$空间的再生核函数为

$\begin{eqnarray*} K_y(x)= \left\{\begin{array}{ll} 1+x-a, & x\leq y,\\ 1+y-a, & x>y. \end{array}\right. \end{eqnarray*}$

下面我们定义一个三次样条空间.

${\bf定义2.3}$${V}_{n,\pi}[a,b]$是一个三次样条空间, 若满足

$ {V}_{n,\pi}[a,b]=\left\{ f_n(x)|f^{(2)}\mbox{在$[a,b]$上是连续的}, f_n|_{\tau_i} \in {\cal P}_3\, i=1,2,\cdots,n \right\}, $

$\pi:a=x_1<x_2<\cdots<x_{n-1}<x_n=b$, $\tau_i=[x_{i-1},x_i]$, $h_i=x_i-x_{i-1}$, $i=1,2,\cdots,n$. 其中, $\pi$是区间$[a,b]$的一个分割, $\tau_i$是小区间, $h_i$是步长, ${\cal P}_k$ 是次数小于等于$k$的多项式构成的线性空间.

接下来, 我们给出空间${V}_{n,\pi}[a,b]$的一组基.

$ {\cal J}_{a}f=\int_{a}^{x} f(t){\rm d}t, \qquad {\cal J}_{a}^{2}f=\int_{a}^{x} {\cal J}_{a}f(t){\rm d}t, $
$ \varphi_i(x)=K_{x_i}(x), $
$ \phi_i(x)={\cal J}_{a}^{2}\varphi_i(t)= \left\{\begin{array}{ll} -\frac{1}{6}(3-a-x)(a-x)^{2}, & x\leq x_i,\\[3mm] \frac{1}{6}(-a^{3}+3a^{2}(1+x)-3ax(2+x)-3xx_{i}^{2}+x_{i}^{3}+3x^{2}(1+x_{i})), & x>x_i. \end{array}\right. $

$i=1,2,\cdots,n$.

${\bf引理2.1}$${V}_{n,\pi}[a,b]=span\{\phi_1,\phi_2,\cdots,\phi_n,1,x-a\}$.

${\bf证}$ 我们只需证明$\phi_1,\phi_2,\cdots,\phi_n,1,x-a$是线性无关的即可.

假设存在$c_1,c_2,\cdots,c_{n+2}$使得

$ \sum_{i=1}^{n}c_i\phi_i(x)+c_{n+1}+c_{n+2}(x-a)=0, $

因为

$ \phi_i(x)={\cal J}_{a}^{2}\varphi_i(t)=\int_{a}^{x} {\cal J}_{a}\varphi_i(t){\rm d}t=\int_{a}^{x}(\int_{a}^{x}\varphi_i(t){\rm d}t){\rm d}t $

两边同时求二阶导可得

$ \sum_{i=1}^{n}c_i\varphi_i(x)=0. $

$l_k(x)$$x_i$的拉格朗日插值基函数, $k=1,2,\cdots,n$, 从而有

$ l_k(x_i)= \left\{\begin{array}{ll} 1,~ & i=k,\\ 0, & i\neq k, \end{array}\right. $

所以

$ 0={\langle 0,l_k(x)\rangle}_{{W}^1_{2}}=\langle \sum_{i=1}^{n}c_i\varphi_i(x),l_k(x)\rangle_{{W}^1_{2}}=\sum_{i=1}^{n}c_{i}\langle K_{x_i}(x),l_k(x)\rangle_{{W}^1_{2}}=\sum_{i=1}^{n}c_{i}l_k(x_i)=c_k. $

进而有$c_{n+1}+c_{n+2}(x-a)=0$.

因为$1$$x-a$线性无关, 所以$c_{n+1}=c_{n+2}=0$, 从而$c_1=c_2=\cdots=c_{n+2}=0$.$\phi_1,\phi_2,\cdots,\phi_n,1,x-a$是线性无关的.

由此我们可知${V}_{n,\pi}[a,b]$的维数为$n+2$.

为了求解方程组(1.1)-(1.3), 我们在$\Omega=(0,1)$上定义一个破裂的三次样条空间.

${\bf定义2.4}$${V}_{N_1,N_2}[0,1]$是破裂的三次样条空间, 若满足

$ {V}_{N_1,N_2}[0,1]=\left\{v(x)|v(x)= \left\{\begin{array}{ll} v_1(x),~ & 0\leq x\leq \alpha,\\ v_2(x), & \alpha<x\leq 1. \end{array}\right. v_1(x) \in {V} _{N_1,\pi_1}[\alpha]\,,v_2(x) \in {V} _{N_2,\pi_2}[\alpha,1]\,\right\}. $

其中

$\begin{eqnarray*} &&\pi_1:0=x_1<x_2<\cdots<x_{n-1}<x_{N_1}=\alpha,\\ &&\pi_2:\alpha=\overline{x_1}<\overline{x_2}<\cdots<\overline{x_{n-1}}<\overline{x_{N_2}}=1 \end{eqnarray*}$

分别是$\Omega_1=(0,\alpha)$$\Omega_2=(\alpha,1)$上的两个分割, $h_i=x_i-x_{i-1},i=1,2,\cdots,N_1$, $\overline{h_j}=\overline{x_j}-\overline{x_{j-1}},j=1,2,\cdots,N_2$分别表示它们的步长, ${V}_{N_1,\pi_1}[\alpha]$, ${V}_{N_2,\pi_2}[\alpha,1]$分别表示$[\alpha]$, $[\alpha,1]$上的三次样条空间.

接下来, 我们给出空间${V}_{N_1,N_2}[0,1]$的一组基.

$\begin{equation} \Phi_i(x)= \left\{\begin{array}{ll} \phi_i(x), ~ & 0\leq x\leq \alpha,\\ 0, & \alpha<x\leq 1, \end{array}\right. \phi_i(x)={\cal J}_{0}^{2}R_{x_i}(x), i=1,2,\cdots,N_1, \end{equation}$
$\begin{equation} \Psi_j(x)= \left\{\begin{array}{ll} 0, & 0\leq x\leq \alpha,\\ \psi_i(x), ~ & \alpha<x\leq 1, \end{array}\right. \psi_i(x)={\cal J}_{\alpha}^{2}r_{x_j}(x), j=1,2,\cdots,N_2, \end{equation} $
$\begin{eqnarray*} \Upsilon_1(x)= \left\{\begin{array}{ll} 1, ~ & x\leq \alpha,\\ 0, & x>\alpha, \end{array}\right.\quad \Upsilon_2(x)= \left\{\begin{array}{ll} x, ~ & x\leq \alpha,\\ 0, & x>\alpha, \end{array}\right.\\ \Upsilon_3(x)= \left\{\begin{array}{ll} 0, ~ & x\leq \alpha,\\ 1, & x>\alpha, \end{array}\right.\quad \Upsilon_4(x)= \left\{\begin{array}{ll} 0, & x\leq \alpha,\\ x-a, ~& x>\alpha. \end{array}\right. \end{eqnarray*}$

其中$R_y(x)$$r_y(x)$分别是$W_2^1[\alpha]$$W_2^1[\alpha,1]$的再生核函数. 从而, 我们得到了一组分段三次样条多项式基函数.

${\bf 引理2.2}$ ${V}_{N_1,N_2}[0,1]=span\{\Phi_1,\Phi_2,\cdots,\Phi_{N_1},\Psi_1,\Psi_2,\cdots,\Psi_{N_2}, \Upsilon_1,\Upsilon_2,\Upsilon_3,\Upsilon_4.$

${\bf证}$ 同引理 2.1, 我们只需证明$\Phi_1,\Phi_2,\cdots,\Phi_{N_1},\Psi_1,\Psi_2,\cdots,\Psi_{N_2}, \Upsilon_1,\Upsilon_2,\Upsilon_3,\Upsilon_4$是线性无关的即可.

假设存在$a_1,a_2,\cdots,a_{N_1},b_1,b_2,\cdots,b_{N_2},c_1,c_2,c_3,c_4$使得

$\begin{eqnarray*} \sum_{i=1}^{N_1}a_i\Phi_i(x)+\sum_{j=1}^{N_2}b_j\Psi_j(x)+\sum_{k=1}^{4}c_k\Upsilon_k(x)=0. \end{eqnarray*}$

$0\leq x\leq \alpha$时, 上式变为

$ \sum_{i=1}^{N_1}a_i\phi_i(x)+c_1+c_2x=0, $

两边同时求二阶导可得

$ \sum_{i=1}^{N_1}a_iR_{x_i}(x)=0. $

由引理$2.1$可知, $a_1=a_2=\cdots=a_{N_1}=c_1=c_2=0$.

同理可得, 当$\alpha\leq x\leq 1$时, $b_1=b_2=\cdots=b_{N_2}=c_3=c_4=0$. 从而有$\Phi_1,\Phi_2,\cdots,\Phi_{N_1},$$ \Psi_1,\Psi_2,\cdots,\Psi_{N_2}, \Upsilon_1,\Upsilon_2,\Upsilon_3,\Upsilon_4$线性无关.

由此可得${V}_{N_1,N_2}[0,1]$的维数为$N_1+N_2+4$.

下面给出样条插值函数的收敛性, 其证明我们不再仔细讨论, 有兴趣的读者可以参考文献[31].

${\bf引理2.3}$$f \in C^{4}[a,b]$, $f_I$$f$$\pi$上的三次样条插值, 则当$h=\max \limits_{i}{{h_i}}\rightarrow 0$时, 对一切$x \in [a,b]$恒有

$\begin{equation}\label{eq:4} |f^{(k)}(x)-f_{I}^{(k)}(x)|\leq c_{k}h^{4-k}, k=0,1,2. \end{equation}$

其中, $c_k$是与分割$\pi$无关的常数.

3 利用最小二乘法求解方程组(1.1)-(1.3)

定义两个线性算子${{\cal L}}_1:C^4[\alpha]\rightarrow C^2[\alpha]$${{\cal L}}_2:C^4[\alpha,1]\rightarrow C^2[\alpha,1]$

$\begin{matrix} &&{\cal L}_1f(x)=-(\beta_1f')'+\gamma_1f, \quad x \in \Omega_1,\end{matrix}$
$\begin{matrix} &&{\cal L}_2f(x)=-(\beta_2f')'+\gamma_2f, \quad x \in \Omega_2. \end{matrix} $

再定义一个线性算子${\cal L}:C^4[0,1] \to C^2[0,1]$

$\begin{eqnarray*} {\cal L}(f)= \left\{\begin{array}{ll} {\cal L}_1(f), ~& x\in \Omega_1,\\ {\cal L}_2(f), & x\in \Omega_2. \end{array}\right. \end{eqnarray*}$

此时, 方程(1.1)-(1.3)等价于如下的算子方程

$\begin{equation}\label{eq:5} \left\{\begin{array}{ll} {\cal L}_1(f)=h_1, \quad x \in \Omega_1,\\ {\cal L}_2(f)=h_2, \quad x \in \Omega_2,\\ f(0)=A,\ f(1)=B,\ \\ {}[f]|_\Omega=\delta_1,\ [\beta f']|_\Omega=\delta_2. \end{array}\right. \end{equation}$

我们在${V}_{N_1,N_2}[0,1]$寻找近似解$f_n^{*}$.

$\begin{eqnarray*} f_n^{*}(x)=\sum_{i=1}^{N_1}a_i\Phi_i(x)+\sum_{j=1}^{N_2}b_j\Psi_j(x)+\sum_{k=1}^{4}c_k\Upsilon_k(x). \end{eqnarray*}$

接下来我们确定未知系数$a_i, b_j, c_k, i=1,2,\cdots,N_1, j=1,2,\cdots,N_2, k=1,2,3,4$.

对任意的$f_n \in {V}_{N_1,N_2}[0,1]$, 设

$\begin{eqnarray*} {\cal D}(f_n)=\left\|{\cal L}f_n-h\right\|_{(0,\Omega)}^2+\left|f_n(0)-A\right|^2+\left|f_n(1) -B\right|^2+\left|[f_n]|_{\Omega}-\delta_1\right|^2+\left| [\beta f_n']|_{\Omega}-\delta_2\right|^2. \end{eqnarray*}$

其中$\left\|{\cal L}f_n-h\right\|_{(0,\Omega)}^2=\left\|{{\cal L}}_1f_n-h_1\right\|_{(0,\Omega_1)}^2+ \left\|{{\cal L}}_2f_n-h_2\right\|_{(0,\Omega_2)}^2$, $\left\|f\right\|_{(0,\Omega_i)}^2=\int_{\Omega_i} f^2(x){\rm d}x$.

$f_n^{*}(x)$应满足

$\begin{equation}\label{eq:6} {\cal D}(f_n^{*})=\min \limits_{f_n\in {V}_{N_1,N_2}[0,1]}{\cal D}(f_n). \end{equation}$

对任意的$f_n,g_n\in {V}_{N_1,N_2}[0,1]$, 定义

$\begin{eqnarray*} {\cal A}(f_n,g_n)=({\cal L}f_n,{\cal L}g_n)_{(0,\Omega)}+f_n(0)g_n(0)+f_n(1)g_n(1)+ [f_n]|_{\Omega}[g_n]|_{\Omega}+[\beta f_n']|_{\Omega}[\beta g_n']|_{\Omega}, \end{eqnarray*} $
$\begin{eqnarray*} \left|f_n\right|_a^2={\cal A}(f_n,f_n). \end{eqnarray*}$

其中$({\cal L}f_n,{\cal L}g_n)_{(0,\Omega)}=({{\cal L}}_1f_n,{{\cal L}}_1g_n)_{(0,\Omega_1)}+ ({{\cal L}}_2f_n,{{\cal L}}_2g_n)_{(0,\Omega_2)}$, $(f,g)_{(0,\Omega_i)}=\int_{\Omega_i} f(x)g(x){\rm d}x$, $\left| \cdot \right|_a$为半范数.

${\bf引理3.1}$$f_n^{*}(x)$满足

$\begin{equation}\label{eq:7} {\cal A}(f_n^{*},g_n)=(h,{{\cal L}}g_n)_{(0,\Omega)}+Ag_n(0)+Bg_n(1)+\delta_1[g_n]|_{\Omega}+\delta_2[\beta g_n']|_{\Omega},\forall v_n \in {V}_{N_1,N_2}[0,1], \end{equation}$

方程(3.5)与(3.4)等价.

${\bf定理3.1}$ 方程(3.5)在${V}_{N_1,N_2}[0,1]$中存在唯一解.

${\bf证}$ 假设$\left|f_n\right|_a=0$, 我们有

$\begin{eqnarray*} {\cal A}(f_n,f_n)=({\cal L}f_n,{\cal L}f_n)_{(0,\Omega)}+f_n(0)f_n(0)+f_n(1)f_n(1)+ [f_n]|_{\Omega}[f_n]|_{\Omega}+[\beta f_n']|_{\Omega}[\beta f_n']|_{\Omega}, \end{eqnarray*}$

所以我们有$\left\|{\cal L}f_n\right\| _{(0,\Omega)}=0$, $f_n(0)=f_n(1)=0$, $[f_n]|_{\Omega}=[\beta f_n']|_{\Omega}=0$.

因此, $f_n$满足方程

$\begin{eqnarray*} \left\{\begin{array}{ll} {\cal L}_1(f)=0, \quad x \in \Omega_1,\\ {\cal L}_2(f)=0, \quad x \in \Omega_2,\\ f(0)=0,\ f(1)=0,\ [f]|_\Omega=0,\ [\beta f']|_\Omega=0. \end{array}\right. \end{eqnarray*}$

算子方程有唯一解$f_n=0$. 所以半范数$\left|f_n\right|_a$${V}_{N_1,N_2}[0,1]$上的范数, 从而方程(3.5)在 ${V} _{N_1,N_2}[0,1]$中存在唯一解.

4 误差分析

${\bf定理4.1}$$f(x)= \left\{\begin{array}{ll} f_1(x), ~ x \in \Omega_1,\\ f_2(x), x \in \Omega_2, \end{array}\right.$ 是方程(3.5)的解, 其中$f_1 \in C^4[\alpha]$, $f_2 \in C^4[\alpha,1]$, 存在常数$c$使得

$\begin{equation} \left|f-f_n^{*}\right|_a\leq ch^2, h=\max \limits_{i,j}\left\{h_i,\overline{h_j}\right\}. \end{equation}$

${\bf证}$ 因为$f(x)$是方程(3.3)的解, 所以有

$\begin{eqnarray*} {\cal A}(f,g_n)&=&({\cal L}f,{\cal L}g_n)_{(0,\Omega)}+f(0)g_n(0)+f(1)g_n(1)+ [f]|_{\Omega}[g_n]|_{\Omega}+[\beta f']|_{\Omega}[\beta g_n']|_{\Omega}\\ &=& (h,{\cal L}g_n)_{(0,\Omega)}+Ag_n(0)+Bg_n(1)+\delta_1[g_n]|_{\Omega}+\delta_2[\beta g_n']|_{\Omega}, \end{eqnarray*}$

$f_I$表示$f$的三次样条插值, 由上式可得

$\begin{eqnarray*} \left|f_n^{*}-f_I\right|_a^2&=&{\cal A}(f_n^{*}-f_I,f_n^{*}-f_I)\\ &=&{\cal A}(f_n^{*},f_n^{*}-f_I)-{\cal A}(f_I,f_n^{*}-f_I)\\ &=&{\cal A}(f,f_n^{*}-f_I)-{\cal A}(f_I,f_n^{*}-f_I)\\ &=&{\cal A}(f-f_I,f_n^{*}-f_I)\\ &\leq &\left|f-f_I\right|_a \cdot \left|f_n^{*}-f_I\right|_a, \end{eqnarray*}$

$ \left|f_n^{*}-f_I\right|_a\leq \left|f-f_I\right|_a. $

由上式和方程(2.3)可知

$ \left|f-f_n^{*}\right|_a\leq ch^2. $

证毕.

类似地, 我们可以在$H^1$$L^2$两个空间中得到以下结论.证明过程运用了有限元Aubin-Nitsche技巧[33].

${\bf定理4.2}$$f(x)= \left\{\begin{array}{ll} f_1(x),~ &x \in \Omega_1,\\ f_2(x), & x \in \Omega_2, \end{array}\right.$ 是方程(3.3)的解, 其中$f_1 \in C^4[\alpha]$, $f_2 \in C^4[\alpha,1]$.存在常数$c$ 使得

$\begin{equation} \left\|f-f_n^{*}\right\|_{1,\Omega}\leq ch^3, h=\max \limits_{i,j}\left\{h_i,\overline{h_j}\right\}. \end{equation}$

$\|f-f_n^{*}\|_{1,\Omega}$$H^1$范数, $\|f-f_n^ {*}\|_{1,\Omega}^2=\|f_1-f_n^{*}\|_{1,\Omega_1}^2+\|f_2-f_n^{*}\|_{1,\Omega_2}^2$.

${\bf证}$ 考虑如下的从属问题

$\begin{eqnarray*} \left\{\begin{array}{ll} {\cal L}_1(F)=f_n-f_I, \quad x \in \Omega_1,\\ {\cal L}_2(F)=f_n-f_I, \quad x \in \Omega_2,\\ F(0)=0,\ F(1)=0,\ [F]|_\Omega=0,\ [\beta F']|_\Omega=0, \end{array}\right. \end{eqnarray*}$

$f_h$为该方程的一个有限元逼近.因为$ F $为方程的精确解, 所以有

$\begin{eqnarray*} ({\cal L}F,{\cal L}(f_h-f_I))_{(0,\Omega)}=(f_h-f_I, {\cal L}(f_n-f_I))_{(0,\Omega)}. \end{eqnarray*}$

进而通过计算,我们有

$\begin{eqnarray*} \|f_n-f_I\|^2_{1,\Omega} \leq Q_1|(f_h-f_I, {\cal L}(f_n-f_I))_{(0,\Omega)}|. \end{eqnarray*}$

因为 $ ({\cal L}F, {\cal L}(f_h-f_I ))_{(0,\Omega)}=({\cal L}F-{\cal L}F_I, {\cal L}(f_h-f_I ))_{(0,\Omega)}+({\cal L}F_I, {\cal L}(f_h-f_I ))_{(0,\Omega)} $, 并且$ {\cal L}(f_h-f_I )={\cal L}(f-f_I )$. 其中

$ | ({\cal L}F-{\cal L}F_I, {\cal L}(f_h-f_I ))_{(0,\Omega)}| \leq \| ({\cal L}F-{\cal L}F_I \|_{(0,\Omega)} \cdot \| {\cal L}(f_h-f_I ) \|_{(0,\Omega)} \leq Q_2h^3\| F\|_{(3,\Omega)}, $
$ |({\cal L}F_I, {\cal L}(f_h-f_I ))_{(0,\Omega)}| \leq Q_2\| F\|_{(3,\Omega)} \cdot \| f-f_I \|_{(1,\Omega)} \leq Q_3h^3\| F\|_{(3,\Omega)}, $

此外,我们有$ \| F\|_{(3,\Omega)} \leq Q_4\|f_n-f_I\|_{(1,\Omega)}$. 综上所述, 定理得证.

${\bf定理4.3}$$f(x)= \left\{\begin{array}{ll} f_1(x),~& x \in \Omega_1,\\ f_2(x), & x \in \Omega_2, \end{array}\right.$ 是方程(3.3)的解, 其中$f_1 \in C^4[\alpha]$, $f_2 \in C^4[\alpha,1]$.存在常数$c$ 使得

$\begin{equation} \left\|f-f_n^{*}\right\|_{(0,\Omega)}\leq ch^4, h=\max \limits_{i,j}\left\{h_i,\overline{h_j}\right\}. \end{equation}$

$\|f-f_n^{*}\|_{(0,\Omega)}$$L^2$范数,$\|f-f_n^ {*}\|_{(0,\Omega)}^2=\|f_1-f_n^{*}\|_{(0,\Omega_1)}^2+\|f_2-f_n^{*}\|_{(0,\Omega_2)}^2.$

${\bf证}$ 考虑如下的从属问题

$ \left\{\begin{array}{ll} {\cal L}_1(G)={\cal J}_{0}^{2}(f_n-f_I), \quad x \in \Omega_1,\\ {\cal L}_2(G)={\cal J}_{a}^{2}(f_n-f_I), \quad x \in \Omega_2,\\ G(0)=0,\ G(1)=0,\ [G]|_\Omega=0,\ [\beta G']|_\Omega=0, \end{array}\right. $

$f_h$为该方程的一个有限元逼近. 证明同定理4.2.

5 数值算例

${\bf算例1}$ 考虑以下界面问题[17]

$ (\beta u_x)_x=-\cos x,\qquad \beta= \left\{\begin{array}{ll} 1,& 0\leq x<\alpha,\\ 5,~& \alpha<x\leq 1, \end{array}\right. $

满足的边界及界面条件为

$ u(0)=1,\qquad u(1)=\frac{1}{5}\cos 1+\frac{2\sqrt{3}}{5},\qquad [u]|_\alpha=0,\qquad [\beta u']|_\alpha=0. $

其中$\alpha=\frac{\pi}{6}$. 问题的精确解为

$ u(x)= \left\{\begin{array}{ll} \cos x,& 0\leq x<\alpha,\\ \frac{1}{5}\cos x+\frac{4}{5}\cos(\alpha),~& \alpha\leq x\leq 1. \end{array}\right. $

算例1利用本文算法所得的近似解如图1所示, 从图1可以看出, 近似解$u_n$能很好的满足精确解$u$. 表1表明了在不同范数下的收敛阶, 这与我们的理论发现是一致的. 表2则是与文献[17]的算法做了比对, 我们的算法精确度更高一些.

图 1

图 1   算例1 $(N_1=N_2=16)$, 左一虚线为近似解图像


表1   算例1在不同范数下误差估计的收敛阶

新窗口打开| 下载CSV


表2   算例1与文献[17]中误差与收敛阶比较

新窗口打开| 下载CSV


表3   算例1中的条件数

新窗口打开| 下载CSV


${\bf算例2}$ 考虑以下界面问题

$ (\beta u_x)_x=30x^4,\qquad \beta= \left\{\begin{array}{ll} 1,~& 0\leq x<\alpha,\\ 4,& \alpha<x\leq 1, \end{array}\right. $

满足的边界及界面条件为

$ u(0)=1,\qquad u(1)=\frac{67}{256},\qquad [u]|_\alpha=0,\qquad [\beta u']|_\alpha=0. $

其中$\alpha=\frac{1}{2}$. 问题的精确解为

$ u(x)= \left\{\begin{array}{ll} x^6,& 0\leq x<\alpha,\\ \frac{1}{4}x^6+\frac{3}{256},~& \alpha\leq x\leq 1. \end{array}\right. $

算例2的近似解与精确解的比较如图2所示. 显然, 用该方法得到的近似解与精确解非常接近. 在表4 中, 给出了$\|\cdot\| _{L^\infty}$, $\|\cdot\|_{L^2}$, $\|\cdot\| _{H^1}$, $\|\cdot\|_{H^2}$$|\cdot|_{a}$范数下的收敛阶.

图 2

图 2   算例2 $(N_1=N_2=16)$, 左一虚线为近似解图像


表4   算例2在不同范数下误差估计的收敛阶

新窗口打开| 下载CSV


表5   算例2中的条件数

新窗口打开| 下载CSV


${\bf算例3}$ 考虑以下界面问题[30]

$ (\beta u_x)_x=f(x),\qquad \beta= \left\{\begin{array}{ll} 1,~& 0\leq x<\alpha,\\ 5,& \alpha<x\leq 1, \end{array}\right. $

满足的边界及界面条件为 $u(0)=0,\ u(1)=e,$$ [u]|_\alpha=e^{\alpha}-\sin(2\pi \alpha),$$ [\beta u']|_\alpha=5e^{\alpha}-2\pi \cos(2\pi \alpha). $ 其中$\alpha=\frac{\pi}{10}$. 问题的精确解为 $ u(x)= \left\{\begin{array}{ll} \sin(2\pi x),~& 0\leq x<\alpha,\\ e^x,& \alpha<x\leq 1. \end{array}\right. $

表6我们可以看出, 与文献[30]中的方法相比, 我们的方法精度更高, 并且在$\|\cdot\|_{L^\infty}$范数下可以达到四阶收敛. 这是优于原文中方法的.

${\bf算例4}$ 考虑以下界面问题[7]

$ (\beta u_x)_x-\gamma u=f(x),\qquad \beta= \left\{\begin{array}{ll} \beta_1(x),~& 0\leq x<\frac{1}{2},\\[2mm] \beta_2(x),& \frac{1}{2}<x\leq 1, \end{array}\right. \qquad \gamma= \left\{\begin{array}{ll} \gamma_1(x),~& 0\leq x<\frac{1}{2},\\[2mm] \gamma_2(x),& \frac{1}{2}<x\leq 1, \end{array}\right. $

其中, $ \beta_1(x)=3e^{-10(x-0.5)^{4}x^{4}},\ \beta_2(x)=x,\ \gamma_1(x)=2,\ \gamma_2(x)=1. $ 问题的精确解为

$ u(x)= \left\{\begin{array}{ll} u_1(x),& 0\leq x<\frac{1}{2},\\[2mm] u_2(x),~& \frac{1}{2}<x\leq 1. \end{array}\right. $

其中, $ u_1(x)=\sin(\pi x),\ u_2(x)=2(x-0.5)^7+1. $

表6   算例3与文献[30]中误差与收敛阶比较

新窗口打开| 下载CSV


图3左表示了近似解与精确解的图像, 其中虚线部分为近似解, 图3右表示绝对误差, 图4左表示精确解的一阶导和近似解的一阶导的误差, 图4右表示精确解的二阶导和近似解的二阶导的误差.

图 3

图 3   算例4 $(N_1=N_2=64)$


图 4

图 4   算例4 $(N_1=N_2=64)$


表7   算例4与文献[7,32]中最大误差值比较

新窗口打开| 下载CSV


表7中,我们将本文的方法与文献[7,32]中提到的方法进行了比较. 可以看出, 我们的方法与文献[7,32]中的方法相比具有更高的精度.

6 结论

在本文中, 我们通过积分$W_2^1$空间的再生核函数, 构造了破裂的三次样条空间的一组基, 然后在我们建立的破裂的三次样条空间上寻找近似解, 运用最小二乘法确定了未知系数并且证明了该方法的收敛性. 最后, 通过几个数值算例印证了我们通过该方法所得的数值解可以很好地与精确解相吻合.

参考文献

Greengard L, Moura M.

On the numerical evaluation of electrostatic fields in composite materials

Acta Numer, 1994, 3: 379-410

DOI:10.1017/S0962492900002464      URL     [本文引用: 1]

A classical problem in electrostatics is the determination of the effective electrical conductivity in a composite material consisting of a collection of piecewise homogeneous inclusions embedded in a uniform background. We discuss recently developed fast algorithms for the evaluation of the potential and electrostatic fields induced in multiphase composites by an applied potential, from which the desired effective properties may be easily obtained. The schemes are based on combining a suitable boundary integral equation with the Fast Multipole Method and the GMRES iterative method; the CPU time required grows linearly with the number of points in the discretization of the interface between the inclusions and the background material.

Peskin C.

Numerical analysis of blood flow in the heart

J Comput Phys, 1977, 25: 220-252

[本文引用: 1]

Al-Smadi M, Arqub O A.

Computational algorithm for solving Fredholm time-fractional partial integrodifferential equations of Dirichlet functions type with error estimates

Appl Math Comput, 2019, 342: 280-294

DOI:10.1016/j.amc.2018.09.020      URL     [本文引用: 1]

Al-Smadi M, Arqub O A, Shawagfeh N, Momani S.

Numerical investigations for systems of second-order periodic boundary value problems using reproducing kernel method

Appl Math Comput, 2016, 291: 137-148

DOI:10.1016/j.amc.2016.06.002      URL     [本文引用: 1]

Albright J, Epshteyn Y, Medvinsky M, Xia Q.

High-order numerical schemes based on difference potentials for 2D elliptic problems with material interfaces

Appl Numer Math, 2017, 111: 64-91

DOI:10.1016/j.apnum.2016.08.017      URL     [本文引用: 2]

Babuska I, Caloz G, Osborn J.

Special finite element methods for a class of second order elliptic problems with rough coefficients

SIAM J Numer Anal, 1994, 31: 945-981

DOI:10.1137/0731051      URL     [本文引用: 1]

Epshteyn Y, Phippen S.

High-order difference potentials methods for 1D elliptic type models

Appl Numer Math, 2015, 93: 69-86

DOI:10.1016/j.apnum.2014.02.005      URL     [本文引用: 5]

Hellrung J, Wang L, Sifakis E, Teran J.

A second order virtual node method for elliptic problems with interfaces and irregular domains in three dimensions

J Comput Phys, 2012, 231: 2015-2048

DOI:10.1016/j.jcp.2011.11.023      URL     [本文引用: 1]

Lee C, Park J.

Preconditioning for finite element methods with strain smoothing

Computers & Mathematics with Applications, 2023, 130: 41-57

DOI:10.1016/j.camwa.2022.11.018      URL     [本文引用: 1]

Kalachikova U, Vasilyeva M, Harris I, et al.

Generalized Multiscale Finite Element Method for scattering problem in heterogeneous media

J Comput Appl Math, 2023, 424: 114977

DOI:10.1016/j.cam.2022.114977      URL     [本文引用: 1]

Guzman J, Sanchez M, Sarkis M.

On the accuracy of finite element approximations to a class of interface problems

Math Comput, 2016, 85: 2071-2098

DOI:10.1090/mcom/2016-85-301      URL     [本文引用: 1]

Harari I, Dolbow J.

Analysis of an efficient finite element method for embedded interface problems

Comput Math, 2010, 46: 205-211

[本文引用: 1]

Dong B Y, Feng X F, Lin Z L.

An FE-FD method for anisotropic elliptic interface problems

SIAM J Sci Comput, 2020, 42: B1041-B1066

DOI:10.1137/19M1291030      URL     [本文引用: 1]

Pan K J, Wu X X, Xu Y F, Yuan G W.

An exact-interface-fitted mesh generator and linearity-preserving finite volume scheme for anisotropic elliptic interface problems

J Comput Phys, 2022, 463: 111293

DOI:10.1016/j.jcp.2022.111293      URL     [本文引用: 1]

Cao W, Zhang X, Zhang Z.

Superconvergence of immersed finite element methods for interface problems

Adv Comput Math, 2017, 43: 795-821

DOI:10.1007/s10444-016-9507-7      URL     [本文引用: 1]

He X, Lin T, Lin Y.

Immersed finite element methods for elliptic interface problems with non-homogeneous jump conditions

Int J Numer Anal Model, 2011, 8: 284-301

[本文引用: 1]

Cao W, Zhang X, Zhang Z, Zou Q.

Superconvergence of immersed finite volume methods for one-dimensional interface problems

J Sci Comput, 2017, 73: 543-565

DOI:10.1007/s10915-017-0532-6      [本文引用: 4]

Liu K, Zou Q.

Analysis of a special immersed finite volume method for elliptic interface problems

Int J Numer Anal Model, 2019, 16: 964-984

[本文引用: 1]

Liu W, Jun S, Zhang Y.

Reproducing kernel particle method

Int J Numer Methods Fluids, 1995, 20: 1081-1106

DOI:10.1002/(ISSN)1097-0363      URL     [本文引用: 1]

Yang J, Guan P, Fan C.

Weighted reproducing kernel collocation method and error analysis for inverse Cauchy problems

Int J Appl Mech, 2016, 8: 1650030

DOI:10.1142/S1758825116500307      URL     [本文引用: 1]

In this work, the weighted reproducing kernel collocation method (weighted RKCM) is introduced to solve the inverse Cauchy problems governed by both homogeneous and inhomogeneous second-order linear partial differential equations. As the inverse Cauchy problem is known for the incomplete boundary conditions, how to numerically obtain an accurate solution to the problem is a challenging task. We first show that the weighted RKCM for solving the inverse Cauchy problems considered is formulated in the least-squares sense. Then, we provide the corresponding error analysis to show how the errors in the domain and on the boundary can be balanced with proper weights. The numerical examples demonstrate that the weighted discrete systems improve the accuracy of solutions and exhibit optimal convergence rates in comparison with those obtained by the traditional direct collocation method. It is shown that neither implementation of regularization nor implementation of iteration is needed to reach the desired accuracy. Further, the locality of reproducing kernel approximation gets rid of the ill-conditioned system.

Yang J, Hsin W.

Weighted reproducing kernel collocation method based on error analysis for solving inverse elasticity problems

Acta Mech, 2019, 230: 3477-3497

DOI:10.1007/s00707-019-02473-0      [本文引用: 1]

Chen J, Wang H, Hu H.

Reproducing kernel enhanced local radial basis collocation method

Int J Numer Methods Eng, 2008, 80: 600-627

[本文引用: 1]

Niu J, Xu M, Yao G M.

An efficient reproducing kernel method for solving the allen-cahn equation

Appl Math Lett, 2019, 89: 78-84

DOI:10.1016/j.aml.2018.09.013      URL     [本文引用: 1]

Niu J, Sun L X, Xu M Q, Hou J J.

A reproducing kernel method for solving heat conduction equations with delay

Appl Math Lett, 2020, 100: 106036

DOI:10.1016/j.aml.2019.106036      URL     [本文引用: 1]

Niu J, Jia Y T, Sun J D.

A new piecewise reproducing kernel function algorithm for solving nonlinear hamiltonian systems

Appl Math Lett, 2023, 136: 108451

DOI:10.1016/j.aml.2022.108451      URL     [本文引用: 1]

Sun L X, Niu J, Hou J J.

A high order convergence collocation method based on the reproducing kernel for general interface problems

Appl Math Lett, 2021, 112: 106718

DOI:10.1016/j.aml.2020.106718      URL     [本文引用: 1]

Yu Y, Niu J, Zhang J, Ning S Y.

A reproducing kernel method for nonlinear C-q-fractional IVPs

Appl Math Lett, 2022, 125: 107751

DOI:10.1016/j.aml.2021.107751      URL     [本文引用: 1]

Yu Y K, Niu J, Yang X M, Cui M G. A broken reproducing kernel method for the multiple interface problems, Comput Appl Math, 2022, 41: 260

[本文引用: 1]

吴勃英, 林迎珍. 应用型再生核空间. 北京: 科学出版社, 2012

[本文引用: 2]

Wu B, Lin Y Z. Applied Regenerative Nuclear Space. Beijing: Science Press, 2012

[本文引用: 2]

杨学敏, 牛晶, 姚春华.

椭圆型界面问题的破裂再生核方法

计算数学, 2022, 44(2): 217-232

DOI:10.12286/jssx.j2021-0791      [本文引用: 3]

本文基于一维椭圆型界面问题提出了一种有效的数值方法.首先,根据模型构建一个崭新的破裂再生核空间.其次,应用破裂再生核方法给出了此类界面问题的近似解,并讨论该方法的收敛性.最后,通过几个有效的数值算例来说明该方法的精确性和稳定性.

Yang X M, Niu J, Yao C H.

The method of fracture regeneration core for elliptical interface problem

Computational Mathematics, 2022, 44(2): 217-232

DOI:10.12286/jssx.j2021-0791      [本文引用: 3]

本文基于一维椭圆型界面问题提出了一种有效的数值方法.首先,根据模型构建一个崭新的破裂再生核空间.其次,应用破裂再生核方法给出了此类界面问题的近似解,并讨论该方法的收敛性.最后,通过几个有效的数值算例来说明该方法的精确性和稳定性.

Hall C A, Meyer W W.

Optimal error bounds for cubic spline interpolation

J Approx Theory, 1976, 16: 105-122

DOI:10.1016/0021-9045(76)90040-X      URL     [本文引用: 1]

Xu M, Zhao Z, Niu J, Lin Y.

A simplified reproducing kernel method for 1-D elliptic type interface problems

J Comput Appl Math, 2019, 351: 29-40

DOI:10.1016/j.cam.2018.10.027      URL     [本文引用: 3]

林群, 谢和虎.

有限元Aubin-Nitsche技巧新认识及其应用

数学的实践与认识, 2011, 41(17): 247-258

[本文引用: 1]

Lin Q, Xie H.

An observation on Aubin-Nitsche lemma and its applications

Mathematics in Practice and Theory, 2011, 41(17): 247-258

[本文引用: 1]

/