主要内容\(\newcommand{\Ima}{\rm Im }
\newcommand{\N}{\mathbb N}
\newcommand{\Z}{\mathbb Z}
\newcommand{\Q}{\mathbb Q}
\newcommand{\R}{\mathbb R}
\newcommand{\F}{\mathbb F}
\newcommand{\C}{\mathbb C}
\newcommand{\K}{\mathbb K}
\newcommand{\myunit}{1 cm}
\newcommand{\blue}[1]{{\color{blue}#1}}
\newcommand\iddots{\mathinner{
\kern1mu\raise1pt{.}
\kern2mu\raise4pt{.}
\kern2mu\raise7pt{\Rule{0pt}{7pt}{0pt}.}
\kern1mu
}}
\tikzset{
node style sp/.style={draw,circle,minimum size=\myunit},
node style ge/.style={circle,minimum size=\myunit},
arrow style mul/.style={draw,sloped,midway,fill=white},
arrow style plus/.style={midway,sloped,fill=white},
}
\newcommand{\lt}{<}
\newcommand{\gt}{>}
\newcommand{\amp}{&}
\definecolor{fillinmathshade}{gray}{0.9}
\newcommand{\fillinmath}[1]{\mathchoice{\colorbox{fillinmathshade}{$\displaystyle \phantom{\,#1\,}$}}{\colorbox{fillinmathshade}{$\textstyle \phantom{\,#1\,}$}}{\colorbox{fillinmathshade}{$\scriptstyle \phantom{\,#1\,}$}}{\colorbox{fillinmathshade}{$\scriptscriptstyle\phantom{\,#1\,}$}}}
\)
节 5.2 正交投影与最小二乘解
练习 练习
基础题.
1.
判断下述向量对是否正交:
-
\(a = \begin{pmatrix}
8\\-5
\end{pmatrix},\ b = \begin{pmatrix}
3\\-1
\end{pmatrix}\);
-
\(u = \begin{pmatrix}
12\\3\\-5
\end{pmatrix},\ v = \begin{pmatrix}
2\\-3\\3
\end{pmatrix}\);
-
\(u = \begin{pmatrix}
3\\2\\-5\\0
\end{pmatrix},\ v = \begin{pmatrix}
-4\\1\\-2\\3
\end{pmatrix}\);
-
\(x = \begin{pmatrix}
-3\\7\\4\\0
\end{pmatrix},\ y = \begin{pmatrix}
1\\-8\\15\\-7
\end{pmatrix}\)。
解答.
-
\(a\cdot b=8\times 3+(-5)\times (-1)=29\neq 0\),所以
\(a,b\)非正交。
-
\(u\cdot v=12 \times 2 + 3\times (-3) + (-5) \times 3 = 0\),故
\(u,v\)正交。
-
\(u \cdot v =3 \times (-4) + 2 \times 1 + (-5)\times (-2) + 0 \times 3 = 0\),故
\(u,v\)正交。
-
\(x \cdot y = (-3) \cdot 1 + 7 \times (-8) + 4 \times 15 + 0 \times (-7) = 1 \neq 0\),所以
\(x,y\)非正交。
2.
求线性方程组\(AX=\beta\)的最小二乘解,其中
\begin{equation*}
A =\begin{pmatrix}
4 & 0\\
0 & 2\\
1 & 1
\end{pmatrix},\quad \beta =\begin{pmatrix}
2\\0\\11
\end{pmatrix}.
\end{equation*}
解答.
因为\(A\)列满秩,所以\(AX=\beta\)有唯一最小二乘解
\begin{equation*}
X=(A^TA)^{-1}A^T\beta=\begin{pmatrix} -10 \\ 2 \end{pmatrix}.
\end{equation*}
提高题.
3.
设\(V_1,V_2\)都是标准内积空间\(\R^n\)的两个子空间。求证:
-
\(\left(V_1^\bot\right)^\bot=V_1\);
-
若
\(V_1\subseteq V_2\),则
\(V_2^\bot\subseteq V_1^\bot\);
-
\(\left(V_1+V_2\right)^\bot=V_1^\bot\cap V_2^\bot\);
-
\(\left(V_1\cap V_2\right)^\bot=V_1^\bot +V_2^\bot\)。
解答.
-
由于
\(\R^n = V_1 \oplus V_1^\bot\),且
\(V_1 \bot V_1^\bot\),所以
\(V_1 = \left(V_1^\bot\right)^\bot\)。
-
对任意
\(\alpha\in V_2^\bot,\beta\in V_1\),由
\(V_1\subseteq V_2\)知
\(\beta\in V_2\),则
\(\alpha \cdot \beta = 0\),故
\(\alpha\in V_1^\bot\)。由
\(\alpha\)的任意性得
\(V_2^\bot\subseteq V_1^\bot\)。
-
对任意\(\alpha\in\left(V_1+V_2\right)^\bot,\beta\in V_1\),由于\(\beta=\beta+{\bf 0}\in V_1+V_2\),所以\(\alpha\cdot\beta=0\),故\(\alpha\in V_1^\bot\)。同理,\(\alpha\in V_2^\bot\)。因此
\begin{equation*}
\left(V_1+V_2\right)^\bot\subseteq V_1^\bot\cap V_2^\bot .
\end{equation*}
另一方面,对任意\(\gamma\in V_1^\bot\bigcap V_2^\bot,\xi\in V_1+V_2\),存在\(\xi_1\in V_1,\xi_2\in V_2\),使得\(\xi=\xi_1+\xi_2\)。由于\(\gamma\cdot\xi_1=\gamma\cdot\xi_2=0\),所以
\begin{equation*}
\gamma\cdot\xi=\gamma\cdot(\xi_1+\xi_2)=\gamma\cdot\xi_1+\gamma\cdot\xi_2=0,
\end{equation*}
即\(\gamma\in \left(V_1+V_2\right)^\bot\),因此\(V_1^\bot\cap V_2^\bot\subseteq\left(V_1+V_2\right)^\bot\)。综上,\(\left(V_1+V_2\right)^\bot=V_1^\bot\cap V_2^\bot\)。
-
\begin{equation*}
\left(V_1^\bot+V_2^\bot\right)^\bot=\left(V_1^\bot\right)^\bot\cap \left(V_2^\bot\right)^\bot=V_1\cap V_2,
\end{equation*}
两边同取正交补得\(\left(V_1\cap V_2\right)^\bot=V_1^\bot +V_2^\bot\)。
4.
设\(U\)是下列齐次线性方程组的解空间:
\begin{equation*}
\left\{\begin{array}{l}
x_1-x_3+x_4=0,\\
x_2+x_3=0,
\end{array}\right.
\end{equation*}
试求:
-
-
解答.
-
\(\displaystyle U^\bot=\langle(1,0,-1,1)^T,(0,1,1,0)^T\rangle.\)
-
因为该齐次线性方程组的基础解系为
\begin{equation*}
\alpha_1=(1,-1,1,0)^T,\alpha_2=(-1,0,0,1)^T,
\end{equation*}
所以\(U^\bot\)适合的线性方程组为
\begin{equation*}
\left\{\begin{array}{l}
x_1-x_2+x_3=0,\\
-x_1+x_4=0,
\end{array}\right.
\end{equation*}
5.
设
\(A\in\mathbb{R}^{m\times n},\beta\in\mathbb{R}^m\),证明:线性方程组
\(AX=\beta\)有解的充分必要条件是
\(\beta\)与
\(A^TX={\bf 0}\)的解空间正交。
解答.
设\(U\)是\(A^TX={\bf 0}\)的解空间,\(A=(\alpha_1,\dots,\alpha_n)\),则对任意\(\beta\in U\),有
\begin{equation*}
\begin{pmatrix}\alpha_1^T \\ \vdots \\ \alpha_n\end{pmatrix}\beta={\bf 0},
\end{equation*}
即
\begin{equation*}
\alpha_i\cdot\beta=\alpha_i^T\beta=0, i=1,\dots ,n,
\end{equation*}
因此\(\alpha_i\in U^\bot,i=1,\dots ,n\),从而\({\rm Im}A\subseteq U^\bot\)。注意到
\begin{equation*}
\dim {\rm Im} A = r(A) = \dim U^\bot,
\end{equation*}
因此\({\rm Im}A = U^\bot\)。
充分性:由于
\(\beta\)与
\(A^TX={\bf 0}\)的解空间正交,
\(\beta\in U^\bot\),因此
\(\beta\in{\rm Im}A\),从而
\(AX=\beta\)有解。
必要性:由于
\(AX=\beta\)有解,所以
\(\beta\in{\rm Im}A\),于是
\(\beta\in U^\bot\),因此
\(\beta\)与
\(A^TX={\bf 0}\)的解空间正交。
挑战题.
6.
设矩阵\(A\in \R^{m\times n}\)满秩分解为\(A=BC\),证明其MP广义逆为
\begin{equation*}
A^{\dagger}=C^{\dagger}B^{\dagger}.
\end{equation*}
举例说明当\(A =BC\)不是满秩分解时,存在\(A^{\dagger}\neq C^{\dagger}B^{\dagger}\)的情形。