数值分析(4)-多项式插值: 埃尔米塔插值法

整理一下数值分析的笔记~
目录：

1. 误差

2. 多项式插值与样条插值(THIS)

3. 函数逼近

4. 数值积分与数值微分

5. 线性方程组的直接解法

6. 线性方程组的迭代解法

7. 非线性方程求根

8. 特征值和特征向量的计算

9. 常微分方程初值问题的数值解

---

1. 埃尔米特插值多项式

定义：设$f(x)$在节点$a\le x_0,x_1,...,x_n\le b$处的函数值为$y_0,y_1,...,y_n$，设$P(x)为f(x)$在区间[a,b]上具有一阶导数的插值函数，

(1) 若要求$P(x)$在区间[a,b]上具有一阶导数(一阶光滑度)，显然$P(x)$在节点$x_0,x_1,...,x_n$处必须满足：

$$\begin{gathered} P(x_i)=f(x_i)=y_i,i=0,1,...,n \\ {P}^{\prime }(x_i)=f^{\prime }(x_i)=y_i^{\prime },i=0,1,...,n \end{gathered}$$

共2n+2个方程解出2n+2个待定的系数，因此P(x)可以是最高次数为2n+1次的多项式，两个节点就是三次多项式作为插值函数。

(2) 若要求P(x)在[a,b]上具有m阶导数(m阶光滑度)，显示P(x)在节点$x_0,x_1,...,x_n$处必须满足：

$$\begin{gathered} P(x_i)=f(x_i)=y_i \\ {P}^{\prime }(x_i)=f^{\prime }(x_i)=y_i^{\prime } \\ {P}^{\prime \prime }(x_i)=f^{\prime \prime }(x_i)=y_i^{\prime \prime },i=0,1,...n \\ ...P^m(x_i)=f^m(x_i)=y_i^m \end{gathered}$$

称满足(1)(2)的插值多项式为诶尔米特插值多项式，记为$H_k(x)$，k为多项式次数。(k过大会影响收敛性和稳定性)

2. 重节点均差

设$f\in C^n[a,b],x_0,x_1,...,x_n$为[a,b]上的相异节点，则$f[x_0,x_1,...,x_n]$是其变量的连续函数，根据均差定义，若$f\in C^1[a,b]$，则有：

$$\begin{gathered} li{m}_{x\to x_0}f[x_0,x]= \\ li{m}_{x\to x_0}\frac{f(x)-f(x_0)}{x-x_0}=f^{\prime }(x_0) \end{gathered}$$

由此定义一阶重节点均差：

$$f[x_0,x_0]=li{m}_{x\to x_0}f[x_0,x]=f^{\prime }(x_0)$$

类似地可以定义重节点的n阶均差：

$$\begin{gathered} f[x_0,x_0,...,x_0(共n个)]= \\ li{m}_{x_i\to x_0}f[x_0,x_1,...,x_n]= \\ \frac{1}{n!}{f}^{(n)}(x_0) \end{gathered}$$

3. 两点三次Hermite插值

$x$	$x_0$	$x_1$
$f(x)$	$y_0$	$y_1$
$f^{\prime }(x)$	$y_0^{\prime }$	$y_1^{\prime }$

两个节点最高可用三次Hermite多项式$H_3(x)$作为插值函数，满足:

$$H_3(x_i)=y_i,H_3^{\prime }(x_i)=y_i^{\prime }(i=0,1)$$

直接设$H_3(x)=a{x}^3+b{x}^2+cx+d$，待定系数使计算复杂，不易推广到高次，可以引入四个基函数：

$${\alpha }_0(x),{\alpha }_1(x),{\beta }_0(x),{\beta }_1(x)$$

使之满足：

$$\begin{gathered} {\alpha }_0(x_0)=1,{\alpha }_1(x_0)=0,{\beta }_0(x_0)=0,{\beta }_1(x_0)=0 \\ {\alpha }_0(x_1)=0,{\alpha }_1(x_1)=1,{\beta }_0(x_1)=0,{\beta }_1(x_1)=0 \\ {\alpha }_0^{\prime }(x_0)=0,{\alpha }_1^{\prime }(x_0)=0,{\beta }_0^{\prime }(x_0)=1,{\beta }_1^{\prime }(x_0)=0 \\ {\alpha }_0^{\prime }(x_1)=0,{\alpha }_1^{\prime }(x_1)=0,{\beta }_0^{\prime }(x_1)=0,{\beta }_1^{\prime }(x_1)=1 \end{gathered}$$

又插值函数应满足：

$$\begin{gathered} H_3(x_0)=y_0,H_3(x_1)=y_1 \\ {H}_3^{\prime }(x_0)=y_0^{\prime },H_3^{\prime }(x_1)=y_1^{\prime } \end{gathered}$$

$H_3(x)$用四个插值基函数表示，假设：

$$\begin{gathered} H_3(x)=y_0{\alpha }_0(x)+y_1{\alpha }_1(x)+y_0^{\prime }{\beta }_0(x)+y_1^{\prime }{\beta }_1(x) \\ {H}_3^{\prime }(x)=y_0{\alpha }_0^{\prime }(x)+y_1{\alpha }_1^{\prime }(x)+y_0^{\prime }{\beta }_0^{\prime }(x)+y_1^{\prime }{\beta }_1^{\prime }(x) \end{gathered}$$

由上面的插值基函数的函数值和导函数值可知：$x_1$是${\alpha }_0(x)$的二重零点{$x_1$处函数值导函数值均为0}，即可假设：

$${\alpha }_0(x)=(x-x_1{)}^2(ax+b)$$

由${\alpha }_0(x_0)=1,{\alpha }_0^{\prime }(x_0)=0$得：

$$\begin{gathered} a=-\frac{2}{(x_0-x_1{)}^3}, \\ b=\frac{1}{(x_0-x_1{)}^2}+\frac{2x_0}{(x_0-x_1{)}^3} \end{gathered}$$

使用类似方法可得：

$$\begin{gathered} {\alpha }_0(x)=(1+2l_1(x))\cdot l_0^2(x)= \\ \left(1+2\frac{x-x_0}{x_1-x_0}\right){\left(\frac{x-x_1}{x_0-x_1}\right)}^2 \end{gathered}$$

$$\begin{gathered} {\alpha }_1(x)=(1+2l_0(x))\cdot l_1^2(x)= \\ \left(1+2\frac{x-x_1}{x_0-x_1}\right){\left(\frac{x-x_0}{x_1-x_0}\right)}^2 \end{gathered}$$

$$\begin{gathered} {\beta }_0(x)=(x-x_0)\cdot l_0^2(x)= \\ (x-x_0){\left(\frac{x-x_1}{x_0-x_1}\right)}^2 \end{gathered}$$

$$\begin{gathered} {\beta }_1(x)=(x-x_1)\cdot l_1^2(x)= \\ (x-x_1){\left(\frac{x-x_0}{x_1-x_0}\right)}^2 \end{gathered}$$

由此可得两点三次Hermite插值公式：

$$\begin{gathered} H_3(x)=y_0{\alpha }_0(x)+y_1{\alpha }_1(x)+y_0^{\prime }{\beta }_0(x)+y_1^{\prime }{\beta }_1(x) \\ {y}_0(1+2l_1(x))\cdot l_0^2(x)+y_1(1+2l_0(x))\cdot l_1^2(x) \\ +y_0^{\prime }(x-x_0)\cdot l_0^2(x)+y_1^{\prime }(x-x_1)\cdot l_1^2(x) \\ =y_0\left(1+2\frac{x-x_0}{x_1-x_0}\right){\left(\frac{x-x_1}{x_0-x_1}\right)}^2 \\ +y_1\left(1+2\frac{x-x_1}{x_0-x_1}\right){\left(\frac{x-x_0}{x_1-x_0}\right)}^2 \\ +y_0^{\prime }(x-x_0){\left(\frac{x-x_1}{x_0-x_1}\right)}^2 \\ +y_1^{\prime }(x-x_1){\left(\frac{x-x_0}{x_1-x_0}\right)}^2 \end{gathered}$$

两点三次Hermite插值的余项:

$$\begin{gathered} R_3(x)=\frac{f^{(4)}(\xi )}{4!}(x-x_0{)}^2(x-x_1{)}^2, \\ {x}_0\le \xi \le x_1 \end{gathered}$$

eg.求作二次多项式使其满足：

$${\phi }_2(0)=y_0,{\phi }_2^{\prime }(0),{\phi }_1(1)=y_1$$

解：

先求基函数${\alpha }_0(x),{\alpha }_1(x),{\beta }_0(x)$满足：

$$\begin{gathered} {\alpha }_0(0)=1,{\alpha }_1(0)=0,{\beta }_0(0)=0 \\ {\alpha }_0(1)=0,{\alpha }_1(1)=1,{\beta }_0(1)=0, \\ {\alpha }_0^{\prime }(0)=0,{\alpha }_1^{\prime }(0)=0,{\beta }_0^{\prime }(0)=1 \end{gathered}$$

可令：

$$\begin{gathered} {\alpha }_0(x)=(a+bx)(x-1) \\ {\alpha }_1(x)=c{x}^2 \\ {\beta }_0(x)=dx(x-1) \end{gathered}$$

可得a=b=-1,c=1,d=-1,则：

$${\phi }_2(x)=y_0(1-x{)}^2+y_1{x}^2+y_0^{\prime }x(1-x)$$

---

{持续更新}
欢迎扫描二维码关注微信公众号 深度学习与数学   [每天获取免费的大数据、AI等相关的学习资源、经典和最新的深度学习相关的论文研读，算法和其他互联网技能的学习，概率论、线性代数等高等数学知识的回顾]