矩阵分解相关知识点总结

〇、Gauss消去法

  对于$n$元线性方程组
$$\mathbf{A}\mathbf{x}=\mathbf{b}\text{\hspace{1em}(0)}$$

其中，$\mathbf{A}=(a_{ij}{)}_{n\times n}$，$\mathbf{x}=({\xi }_1,{\xi }_2,\cdots ,{\xi }_n{)}^{\mathrm{T}}$，$\mathbf{b}=(b_1,b_2,\cdots ,b_n{)}^{\mathrm{T}}$。Gauss消去法的基本思想是化系数矩阵$\text{A}$为上三角矩阵，或化增广矩阵$\left[\begin{array}{cc}\mathbf{A}& \mathbf{b}\end{array}\right]$为上阶梯形矩阵以求其解，这个过程就叫Gauss消元。

  Gauss消元过程能够进行到底的条件是当且仅当$\mathbf{A}$的各阶顺序主子式都不为零，即：
$${\Delta }_r̸=0(r=1,2,\cdots ,n-1)$$

一、矩阵的三角分解

$$三角分解\{\begin{array}{ll}A=LDU& \text{LDU分解}\\ A=L(DU)=L\hat{U}& \text{Doolittle分解}\\ A=(LD)U=\hat{L}U& \text{Crout分解}\end{array}$$

  三角分解又称作LU分解或LR分解。在上式中，$L$是单位下三角矩阵，$D$是对角矩阵，$U$是单位上三角矩阵。

分解过程：
$$A=A^{(0)}=L_1{A}^{(1)}=L_1{L}_2{A}^{(2)}=\cdots =L_1{L}_2\cdots L_{n-1}{A}^{(n-1)}=L{A}^{(n-1)}\text{\hspace{1em}(1)}$$

$L_1=\left[\begin{array}{cccc}1& & & \\ c_{21}& 1& & \\ ⋮& & \ddots & \\ c_{n1}& & & 1\end{array}\right]$，$L_2=\left[\begin{array}{ccccc}1& & & & \\ & 1& & & \\ & c_{32}& 1& & \\ & ⋮& & \ddots & \\ & c_{n2}& & & 1\end{array}\right]$，$\cdots$，$L_r=\left[\begin{array}{cccccc}1& & & & & \\ & \ddots & & & & \\ & & 1& & & \\ & & c_{r+1,r}& 1& & \\ & & ⋮& & \ddots & \\ & & c_{nr}& & & 1\end{array}\right]$，$L_{n-1}=\left[\begin{array}{ccccc}1& & & & \\ & 1& & & \\ & & \ddots & & \\ & & & 1& \\ & & & c_{n,n-1}& 1\end{array}\right]$为Frobenius矩阵，其中空白处全为$0$，$c_{ir}=\frac{a_{ir}^{(r-1)}}{a_{rr}^{(r-1)}},(r=1,2,\cdots ,n-1),(i=r+1,r+2,\cdots ,n)$，$L_r^{-1}=\left[\begin{array}{cccccc}1& & & & & \\ & \ddots & & & & \\ & & 1& & & \\ & & -c_{r+1,r}& 1& & \\ & & ⋮& & \ddots & \\ & & -c_{nr}& & & 1\end{array}\right]$。

$L=L_1{L}_2\cdots L_{(n-1)}=\left[\begin{array}{ccccc}1& & & & \\ c_{21}& 1& & & \\ ⋮& ⋮& \ddots & & \\ c_{n-1,1}& c_{n-1,2}& \cdots & 1& \\ c_{n1}& c_{n2}& \cdots & c_{n,n-1}& 1\end{array}\right]$是一个单位下三角矩阵，$A^{(n-1)}=L_{n-1}^{-1}\cdots L_2^{-1}{L}_1^{-1}{A}^{(0)}=L^{-1}A$是一个上三角矩阵，可以分解成一个对角矩阵（$D$）和一个单位上三角矩阵（$U$）。

二、矩阵的Cholesky分解

  当$A$为实对称正定矩阵时，有
$$A=LDU=A^{\text{T}}=(LDU{)}^{\mathrm{T}}=U^{\mathrm{T}}D{L}^{\mathrm{T}},D=\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}(d_1,d_2,\cdots ,d_n)$$

即有：$L=U^{\mathrm{T}}$，$U=L^{\mathrm{T}}$，令$\tilde{D}=\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}(\sqrt{d_1},\sqrt{d_2},\cdots ,\sqrt{d_n})$，则有：
$$A=LD{L}^{\mathrm{T}}=L{\tilde{D}}^2{L}^{\mathrm{T}}=(L\tilde{D})(\tilde{D}{L}^{\mathrm{T}})=(L\tilde{D})(L\tilde{D}{)}^{\mathrm{T}}=G{G}^{\mathrm{T}}\text{\hspace{1em}(2)}$$

其中，$G=L\tilde{D}$为下三角矩阵。上式即为矩阵的Cholesky分解，又称平方根分解或对称三角分解。

  Cholesky分解过程中，可以根据以下递推公式计算$G$矩阵中每个元素的值：
$$g_{ij}=\{\begin{array}{ll}(a_{ii}-\sum _{k=1}^{i-1}{g}_{ik}^2{)}^{1/2}& (i=j)\\ \frac{1}{g_{jj}}(a_{ij}-\sum _{k=1}^{j-1}{g}_{ik}{g}_{jk})& (i>j)\\ 0& (i<j)\end{array}$$

当然也可以先对矩阵$A$做LDU分解，然后取$G=L\tilde{D}$。

三、矩阵的QR分解

3.1、Givens矩阵与Givens变换

  设非零列向量$\mathbf{x}\in {\mathbf{R}}^n$及单位列向量$\mathbf{z}\in {\mathbf{R}}^n$，存在有限个Givens矩阵的乘积，记作$\mathbf{T}$，使得
$$\mathbf{T}\mathbf{x}=|\mathbf{x}|\mathbf{z}\text{\hspace{1em}(3)}$$

上式即为Givens变换，也称初等旋转变换，其中Givens矩阵，也称初等旋转矩阵，记作${\mathbf{T}}_{ij}={\mathbf{T}}_{ij}(c,s)=\left[\begin{array}{ccccc}\mathbf{I}& & & & \\ & c& & s& \\ & & \mathbf{I}& & \\ & -s& & c& \\ & & & & \mathbf{I}\end{array}\right]$，$\mathbf{T}={\mathbf{T}}_{1n}{\mathbf{T}}_{1,n-1}\cdots {\mathbf{T}}_{13}{\mathbf{T}}_{12}$。

  对于非零列向量$\mathbf{x}=({\xi }_1,{\xi }_2,\cdots ,{\xi }_n{)}^{\mathrm{T}}$，及单位列向量$\mathbf{z}={\mathbf{e}}_1=(1,0,\cdots ,0{)}^{\mathrm{T}}$，其Givens变换过程如下：

• 首先对$\mathbf{x}$构造Givens矩阵${\mathbf{T}}_{12}(c,s)=\left[\begin{array}{ccc}c& s& \\ -s& c& \\ & & \mathbf{I}\end{array}\right]$，其中$c=\frac{{\xi }_1}{\sqrt{{\xi }_1^2+{\xi }_2^2}},s=\frac{{\xi }_2}{\sqrt{{\xi }_1^2+{\xi }_2^2}}$，有
  $${\mathbf{T}}_{12}\mathbf{x}=(\sqrt{{\xi }_1^2+{\xi }_2^2},0,{\xi }_3,\cdots ,{\xi }_n{)}^{\mathrm{T}}$$

• 再对${\mathbf{T}}_{12}\mathbf{x}$构造Givens矩阵${\mathbf{T}}_{13}(c,s)=\left[\begin{array}{cccc}c& & s& \\ & 1& & \\ -s& & c& \\ & & & \mathbf{I}\end{array}\right]$，其中$c=\frac{\sqrt{{\xi }_1^2+{\xi }_2^2}}{\sqrt{{\xi }_1^2+{\xi }_2^2+{\xi }_3^2}},s=\frac{{\xi }_3}{\sqrt{{\xi }_1^2+{\xi }_2^2+{\xi }_3^2}}$，有
  $${\mathbf{T}}_{13}({\mathbf{T}}_{12}\mathbf{x})=(\sqrt{{\xi }_1^2+{\xi }_2^2+{\xi }_3^2},0,0,{\xi }_4,\cdots ,{\xi }_n{)}^{\mathrm{T}}$$

• 如此下去，最后对${\mathbf{T}}_{1,n-1}{\mathbf{T}}_{1,n-2}\cdots {\mathbf{T}}_{13}{\mathbf{T}}_{12}\mathbf{x}$构造Givens矩阵${\mathbf{T}}_{1n}(c,s)=\left[\begin{array}{ccc}c& & s\\ & \mathbf{I}& \\ -s& & c\end{array}\right]$，其中$c=\frac{\sqrt{{\xi }_1^2+\cdots +{\xi }_{n-1}^2}}{\sqrt{{\xi }_1^2+{\xi }_2^2+\cdots +{\xi }_{n-1}^2+{\xi }_n^2}},s=\frac{{\xi }_n}{\sqrt{{\xi }_1^2+{\xi }_2^2+\cdots +{\xi }_{n-1}^2+{\xi }_n^2}}$，有
  $${\mathbf{T}}_{1n}({\mathbf{T}}_{1,n-1}\cdots {\mathbf{T}}_{12}\mathbf{x})=(\sqrt{{\xi }_1^2+{\xi }_2^2+\cdots +{\xi }_{n-1}^2+{\xi }_n^2},0,\cdots ,0{)}^{\mathrm{T}}$$

令$\mathbf{T}={\mathbf{T}}_{1n}{\mathbf{T}}_{1,n-1}\cdots {\mathbf{T}}_{12}$，有$\mathbf{T}\mathbf{x}=|\mathbf{x}|\mathbf{z}=|\mathbf{x}|{\mathbf{e}}_1$，即通过有限个Givens矩阵$\mathbf{T}$将$\mathbf{x}$变换为与$\mathbf{z}$同方向的向量。

3.2、Householder矩阵与Householder变换

  任意给定非零列向量$\mathbf{x}\in {\mathbf{R}}^n(n>1)$及单位列向量$\mathbf{z}\in {\mathbf{R}}^n$，则存在矩阵$\mathbf{H}$，使得
$$\mathbf{H}=|\mathbf{x}|\mathbf{z}\text{\hspace{1em}(4)}$$

上式即为Householder变换，也称初等反射变换，其中$\mathbf{H}=\mathbf{I}-2{\mathbf{u}\mathbf{u}}^{\mathrm{T}}$，为Householder矩阵，也称初等反射矩阵。

  对于非零列向量$\mathbf{x}=({\xi }_1,{\xi }_2,\cdots ,{\xi }_{n-1},{\xi }_n{)}^{\mathrm{T}}$及单位列向量$\mathbf{z}={\text{e}}_1=(1,0,\cdots ,0{)}^{\mathrm{T}}$，其Householder变换过程如下：

  取$\mathbf{u}=\frac{\mathbf{x}-|\mathbf{x}|\mathbf{z}}{|\mathbf{x}-|\mathbf{x}|\mathbf{z}|}=\frac{\mathbf{x}-|\mathbf{x}|{\mathbf{e}}_1}{|\mathbf{x}-|\mathbf{x}|{\mathbf{e}}_1|}$，其中$|\mathbf{x}|=\sqrt{{\xi }_1^2+{\xi }_2^2+\cdots +{\xi }_{n-1}^2+{\xi }_n^2}$，则$\mathbf{H}=\mathbf{I}-2{\mathbf{u}\mathbf{u}}^{\mathrm{T}}$，$\mathbf{H}\mathbf{x}=|\mathbf{x}|\mathbf{z}=|\mathbf{x}|{\mathbf{e}}_1$，即通过Householder矩阵$\mathbf{H}$将$\mathbf{x}$变换为与$\mathbf{z}$同方向的向量。

	Givens矩阵${\text{T}}_{ij}$具有如下性质	Householder矩阵$\text{H}$具有如下性质
(1)	$${\mathbf{T}}_{ij}=-{\mathbf{T}}_{ij}^{\mathrm{T}}=-{\mathbf{T}}_{ij}^{-1}$$	$$\mathbf{H}={\mathbf{H}}^{\mathrm{T}}={\mathbf{H}}^{-1}$$
(2)	$${\mathbf{T}}_{ij}^2=-{\mathbf{T}}_{ij}^{\mathrm{T}}{\mathbf{T}}_{ij}=-{\mathbf{T}}_{ij}^{-1}{\mathbf{T}}_{ij}=-\mathbf{I}$$	$${\mathbf{H}}^2={\mathbf{H}}^{\mathrm{T}}\mathbf{H}={\mathbf{H}}^{-1}\mathbf{H}=\mathbf{I}$$
(3)	$$\mathrm{d}\mathrm{e}\mathrm{t}{\mathbf{T}}_{\mathrm{i}\mathrm{j}}=1$$	$$\mathrm{d}\mathrm{e}\mathrm{t}\mathbf{H}=-1$$

$$初等旋转矩阵是两个初等反射矩阵的乘积，即有{\mathbf{T}}_{ij}={\mathbf{H}}_v{\mathbf{H}}_u\text{\hspace{1em}(5)}$$

3.3、QR分解

  设$A$是$m\times n$实（复）矩阵，且其$n$个列线性无关，则$A$有分解
$$A=QR\text{\hspace{1em}(6)}$$

其中$Q$是$m\times n$实（复）矩阵，且满足$Q^{\text{T}}Q=I$（$Q^{\text{H}}Q=I$），$R$是$n$阶实（复）可逆上三角矩阵。上式即为矩阵的QR分解，也称正交三角分解，该分解除去相差一个对角元素的绝对值（模）全等于1的对角矩阵因子外是唯一的。

  对于任意的$n$阶实可逆矩阵$A=(a_{ij}{)}_{n\times n}$，均可通过左连乘Givens矩阵（初等旋转矩阵）或左连乘Householder矩阵（初等反射矩阵），将其化为可逆上三角矩阵。

  完成矩阵的QR分解有三种常用的方法：Schmidt正交化方法，Givens变换方法和Householder变换方法。下面通过一个具体实例，将矩阵$A=\left[\begin{array}{ccc}1& 2& 2\\ 2& 1& 2\\ 1& 2& 1\end{array}\right]$分别用上述三种方法进行QR分解。

Schmidt正交化方法：

  令${\mathbf{a}}_1=(1,2,1{)}^{\mathrm{T}},{\mathbf{a}}_2=(2,1,2{)}^{\mathrm{T}},{\mathbf{a}}_3=(2,2,1{)}^{\mathrm{T}}$，将其Schmidt正交化可得：
$$\begin{array}{l}{\mathbf{b}}_1={\mathbf{a}}_1=(1,2,1{)}^{\mathrm{T}}\\ {\mathbf{b}}_2={\mathbf{a}}_2-\frac{({\mathbf{a}}_2,{\mathbf{b}}_1)}{({\mathbf{b}}_1,{\mathbf{b}}_1)}{\mathbf{b}}_1={\mathbf{a}}_2-{\mathbf{b}}_1=(1,-1,1{)}^{\mathrm{T}}\\ {\mathbf{b}}_3={\mathbf{a}}_3-\frac{({\mathbf{a}}_3,{\mathbf{b}}_1)}{({\mathbf{b}}_1,{\mathbf{b}}_1)}{\mathbf{b}}_1-\frac{({\mathbf{a}}_3,{\mathbf{b}}_2)}{({\mathbf{b}}_2,{\mathbf{b}}_2)}{\mathbf{b}}_2={\mathbf{a}}_3-\frac{1}{3}{\mathbf{b}}_2-\frac{7}{6}{\mathbf{b}}_1=(\frac{1}{2},0,-\frac{1}{2}{)}^{\mathrm{T}}\end{array}$$

进而有：
$$({\mathbf{a}}_1,{\mathbf{a}}_2,{\mathbf{a}}_3)=({\mathbf{b}}_1,{\mathbf{b}}_2,{\mathbf{b}}_3)\cdot C=({\mathbf{b}}_1,{\mathbf{b}}_2,{\mathbf{b}}_3)\left[\begin{array}{ccc}1& 1& \frac{7}{6}\\ 0& 1& \frac{1}{3}\\ 0& 0& 1\end{array}\right]$$

  取：
$$\begin{array}{l}{\mathbf{q}}_1=\frac{1}{|{\mathbf{b}}_1|}|{\mathbf{b}}_1|=\frac{1}{\sqrt{6}}(1,2,1{)}^{\mathrm{T}}\\ {\mathbf{q}}_2=\frac{1}{|{\mathbf{b}}_2|}|{\mathbf{b}}_2|=\frac{1}{\sqrt{3}}(1,-1,1{)}^{\mathrm{T}}\\ {\mathbf{q}}_3=\frac{1}{|{\mathbf{b}}_3|}|{\mathbf{b}}_3|=\frac{1}{\sqrt{2}}(1,0,-1{)}^{\mathrm{T}}\end{array}$$

  令：
$$\begin{array}{l}Q=({\mathbf{q}}_1,{\mathbf{q}}_2,{\mathbf{q}}_3)=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{2}}\\ \frac{2}{\sqrt{6}}& -\frac{1}{\sqrt{3}}& 0\\ \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}& -\frac{1}{\sqrt{2}}\end{array}\right]\\ R=\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}(|{\mathbf{b}}_1|,|{\mathbf{b}}_2|,|{\mathbf{b}}_3|)\cdot \mathrm{C}=\left[\begin{array}{ccc}\sqrt{6}& 0& 0\\ 0& \sqrt{3}& 0\\ 0& 0& \sqrt{2}\end{array}\right]\left[\begin{array}{ccc}1& 1& \frac{7}{6}\\ 0& 1& \frac{1}{3}\\ 0& 0& 1\end{array}\right]=\left[\begin{array}{ccc}\sqrt{6}& \sqrt{6}& \frac{7}{\sqrt{6}}\\ 0& \sqrt{3}& \frac{1}{\sqrt{3}}\\ 0& 0& \frac{1}{\sqrt{2}}\end{array}\right]\end{array}$$

则有$A=QR$。

Givens变换方法：
  第1步，对$A^{(0)}=A=\left[\begin{array}{ccc}1& 2& 2\\ 2& 1& 2\\ 1& 2& 1\end{array}\right]$的第1列${\mathbf{b}}^{(1)}=\left[\begin{array}{c}1\\ 2\\ 1\end{array}\right]$构造旋转矩阵$T_1$，使$T_1{\mathbf{b}}^{(1)}=|{\mathbf{b}}^{(1)}|{\mathbf{e}}_1$：

$$T_1=T_{13}{T}_{12}=\left[\begin{array}{ccc}\frac{\sqrt{5}}{\sqrt{6}}& 0& \frac{1}{\sqrt{6}}\\ 0& 1& 0\\ -\frac{1}{\sqrt{6}}& 0& \frac{\sqrt{5}}{\sqrt{6}}\end{array}\right]\left[\begin{array}{ccc}\frac{1}{\sqrt{5}}& \frac{2}{\sqrt{5}}& 0\\ -\frac{2}{\sqrt{5}}& \frac{1}{\sqrt{5}}& 0\\ 0& 0& 1\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ -\frac{2}{\sqrt{5}}& \frac{1}{\sqrt{5}}& 0\\ -\frac{1}{\sqrt{30}}& -\frac{2}{\sqrt{30}}& \frac{\sqrt{5}}{\sqrt{6}}\end{array}\right]$$

$$T_1{\mathbf{b}}^{(1)}=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ -\frac{2}{\sqrt{5}}& \frac{1}{\sqrt{5}}& 0\\ -\frac{1}{\sqrt{30}}& -\frac{2}{\sqrt{30}}& \frac{\sqrt{5}}{\sqrt{6}}\end{array}\right]\left[\begin{array}{c}1\\ 2\\ 1\end{array}\right]=\left[\begin{array}{c}\sqrt{6}\\ 0\\ 0\end{array}\right]$$

$$T_1{A}^{(0)}=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ -\frac{2}{\sqrt{5}}& \frac{1}{\sqrt{5}}& 0\\ -\frac{1}{\sqrt{30}}& -\frac{2}{\sqrt{30}}& \frac{\sqrt{5}}{\sqrt{6}}\end{array}\right]\left[\begin{array}{ccc}1& 2& 2\\ 2& 1& 2\\ 1& 2& 1\end{array}\right]=\left[\begin{array}{ccc}\sqrt{6}& \sqrt{6}& \frac{7}{\sqrt{6}}\\ 0& -\frac{3}{\sqrt{5}}& -\frac{2}{\sqrt{5}}\\ 0& \frac{6}{\sqrt{30}}& -\frac{1}{\sqrt{30}}\end{array}\right]$$

  第2步，对$A^{(1)}=\left[\begin{array}{cc}-\frac{3}{\sqrt{5}}& -\frac{2}{\sqrt{5}}\\ \frac{6}{\sqrt{30}}& -\frac{1}{\sqrt{30}}\end{array}\right]$的第1列${\mathbf{b}}^{(2)}=\left[\begin{array}{c}-\frac{3}{\sqrt{5}}\\ \frac{6}{\sqrt{30}}\end{array}\right]$构造旋转矩阵$T_2$，使$T_2{\mathbf{b}}^{(2)}=|{\mathbf{b}}^{(2)}|{\mathbf{e}}_1$：

$$T_2=T_{12}=\left[\begin{array}{cc}-\frac{\sqrt{3}}{\sqrt{5}}& \frac{2}{\sqrt{10}}\\ -\frac{2}{\sqrt{10}}& -\frac{\sqrt{3}}{\sqrt{5}}\end{array}\right]$$

$$T_2{\mathbf{b}}^{(2)}=\left[\begin{array}{cc}-\frac{\sqrt{3}}{\sqrt{5}}& \frac{2}{\sqrt{10}}\\ -\frac{2}{\sqrt{10}}& -\frac{\sqrt{3}}{\sqrt{5}}\end{array}\right]\left[\begin{array}{c}-\frac{3}{\sqrt{5}}\\ \frac{6}{\sqrt{30}}\end{array}\right]=\left[\begin{array}{c}\sqrt{3}\\ 0\end{array}\right]$$

$$T_2{A}^{(1)}=\left[\begin{array}{cc}-\frac{\sqrt{3}}{\sqrt{5}}& \frac{2}{\sqrt{10}}\\ -\frac{2}{\sqrt{10}}& -\frac{\sqrt{3}}{\sqrt{5}}\end{array}\right]\left[\begin{array}{cc}-\frac{3}{\sqrt{5}}& -\frac{2}{\sqrt{5}}\\ \frac{6}{\sqrt{30}}& -\frac{1}{\sqrt{30}}\end{array}\right]=\left[\begin{array}{cc}\sqrt{3}& \frac{1}{\sqrt{3}}\\ 0& \frac{1}{\sqrt{2}}\end{array}\right]$$

  最后，令
$$T=\left[\begin{array}{cc}1& O^{\text{T}}\\ O& T_2\end{array}\right]T_1=\left[\begin{array}{ccc}1& 0& 0\\ 0& -\frac{\sqrt{3}}{\sqrt{5}}& \frac{2}{\sqrt{10}}\\ 0& -\frac{2}{\sqrt{10}}& -\frac{\sqrt{3}}{\sqrt{5}}\end{array}\right]\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ -\frac{2}{\sqrt{5}}& \frac{1}{\sqrt{5}}& 0\\ -\frac{1}{\sqrt{30}}& -\frac{2}{\sqrt{30}}& \frac{\sqrt{5}}{\sqrt{6}}\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ \frac{1}{\sqrt{3}}& -\frac{1}{\sqrt{3}}& \frac{1}{\sqrt{3}}\\ \frac{1}{\sqrt{2}}& 0& -\frac{1}{\sqrt{2}}\end{array}\right]$$

$$Q=T^{\text{T}}==\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{2}}\\ \frac{2}{\sqrt{6}}& -\frac{1}{\sqrt{3}}& 0\\ \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}& -\frac{1}{\sqrt{2}}\end{array}\right],R=TA=\left[\begin{array}{ccc}\sqrt{6}& \sqrt{6}& \frac{7}{\sqrt{6}}\\ 0& \sqrt{3}& \frac{1}{\sqrt{3}}\\ 0& 0& \frac{1}{\sqrt{2}}\end{array}\right]$$

则有$A=QR$。

Householder变换方法：
  第1步，对$A^{(0)}=A=\left[\begin{array}{ccc}1& 2& 2\\ 2& 1& 2\\ 1& 2& 1\end{array}\right]$的第1列${\mathbf{b}}^{(1)}=\left[\begin{array}{c}1\\ 2\\ 1\end{array}\right]$构造Householder矩阵$H_1$，使$H_1{\mathbf{b}}^{(1)}=|{\mathbf{b}}^{(1)}|{\mathbf{e}}_1$：
$${\mathbf{b}}^{(1)}-|{\mathbf{b}}^{(1)}|{\mathbf{e}}_1=\left[\begin{array}{c}1\\ 2\\ 1\end{array}\right]-\sqrt{6}\left[\begin{array}{c}1\\ 0\\ 0\end{array}\right]=\left[\begin{array}{c}1-\sqrt{6}\\ 2\\ 1\end{array}\right]$$

$$\mathbf{u}=\frac{{\mathbf{b}}^{(1)}-|{\mathbf{b}}^{(1)}|{\mathbf{e}}_1}{|{\mathbf{b}}^{(1)}-|{\mathbf{b}}^{(1)}|{\mathbf{e}}_1|}=\frac{1}{\sqrt{12-2\sqrt{6}}}\left[\begin{array}{c}1-\sqrt{6}\\ 2\\ 1\end{array}\right]$$

$$H_1=I-2{\mathbf{u}\mathbf{u}}^{\mathrm{T}}=\left[\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& 1\end{array}\right]-\frac{1}{6-\sqrt{6}}\left[\begin{array}{c}1-\sqrt{6}\\ 2\\ 1\end{array}\right]\left[\begin{array}{ccc}1-\sqrt{6}& 2& 1\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ \frac{2}{\sqrt{6}}& \frac{\sqrt{6}-2}{\sqrt{6}-6}& \frac{2}{\sqrt{6}-6}\\ \frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}-6}& \frac{\sqrt{6}-5}{\sqrt{6}-6}\end{array}\right]$$

$$H_1{A}^{(0)}=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ \frac{2}{\sqrt{6}}& \frac{\sqrt{6}-2}{\sqrt{6}-6}& \frac{2}{\sqrt{6}-6}\\ \frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}-6}& \frac{\sqrt{6}-5}{\sqrt{6}-6}\end{array}\right]\left[\begin{array}{ccc}1& 2& 2\\ 2& 1& 2\\ 1& 2& 1\end{array}\right]=\left[\begin{array}{ccc}\sqrt{6}& \sqrt{6}& \frac{7}{\sqrt{6}}\\ 0& \frac{3-\sqrt{6}}{\sqrt{6}-1}& \frac{2}{\sqrt{6}}\\ 0& \frac{\sqrt{6}}{\sqrt{6}-1}& \frac{1}{\sqrt{6}}\end{array}\right]$$

  第2步，对$A^{(1)}=\left[\begin{array}{cc}\frac{3-\sqrt{6}}{\sqrt{6}-1}& \frac{2}{\sqrt{6}}\\ \frac{\sqrt{6}}{\sqrt{6}-1}& \frac{1}{\sqrt{6}}\end{array}\right]$的第1列${\mathbf{b}}^{(2)}=\left[\begin{array}{c}\frac{3-\sqrt{6}}{\sqrt{6}-1}\\ \frac{\sqrt{6}}{\sqrt{6}-1}\end{array}\right]$构造Householder矩阵$H_2$，使$H_2{\mathbf{b}}^{(2)}=|{\mathbf{b}}^{(2)}|{\mathbf{e}}_1$：
$${\mathbf{b}}^{(2)}-|{\mathbf{b}}^{(2)}|{\mathbf{e}}_1=\left[\begin{array}{c}\frac{3-\sqrt{6}}{\sqrt{6}-1}\\ \frac{\sqrt{6}}{\sqrt{6}-1}\end{array}\right]-\frac{\sqrt{21-6\sqrt{6}}}{\sqrt{6}-1}\left[\begin{array}{c}1\\ 0\end{array}\right]=\left[\begin{array}{c}\frac{3-\sqrt{6}-\sqrt{21-6\sqrt{6}}}{\sqrt{6}-1}\\ \frac{\sqrt{6}}{\sqrt{6}-1}\end{array}\right]=\left[\begin{array}{c}\frac{x}{\sqrt{6}-1}\\ \frac{\sqrt{6}}{\sqrt{6}-1}\end{array}\right]$$

$$\mathbf{u}=\frac{{\mathbf{b}}^{(2)}-|{\mathbf{b}}^{(2)}|{\mathbf{e}}_1}{|{\mathbf{b}}^{(2)}-|{\mathbf{b}}^{(2)}|{\mathbf{e}}_1|}=\frac{\sqrt{6}-1}{\sqrt{x^2+6}}\left[\begin{array}{c}\frac{x}{\sqrt{6}-1}\\ \frac{\sqrt{6}}{\sqrt{6}-1}\end{array}\right]=\left[\begin{array}{c}\frac{x}{\sqrt{x^2+6}}\\ \frac{\sqrt{6}}{\sqrt{x^2+6}}\end{array}\right]$$

$$H_2=I-2{\mathbf{u}\mathbf{u}}^{\text{T}}=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]-2\left[\begin{array}{c}\frac{x}{\sqrt{x^2+6}}\\ \frac{\sqrt{6}}{\sqrt{x^2+6}}\end{array}\right]\left[\begin{array}{cc}\frac{x}{\sqrt{x^2+6}}& \frac{\sqrt{6}}{\sqrt{x^2+6}}\end{array}\right]=\left[\begin{array}{cc}\frac{6-x^2}{x^2+6}& \frac{-2\sqrt{6}x}{x^2+6}\\ \frac{-2\sqrt{6}x}{x^2+6}& \frac{x^2-6}{x^2+6}\end{array}\right]$$

$$H_2{A}^{(1)}=\left[\begin{array}{cc}\frac{6-x^2}{x^2+6}& \frac{-2\sqrt{6}x}{x^2+6}\\ \frac{-2\sqrt{6}x}{x^2+6}& \frac{x^2-6}{x^2+6}\end{array}\right]\left[\begin{array}{cc}\frac{3-\sqrt{6}}{\sqrt{6}-1}& \frac{2}{\sqrt{6}}\\ \frac{\sqrt{6}}{\sqrt{6}-1}& \frac{1}{\sqrt{6}}\end{array}\right]=\left[\begin{array}{cc}\sqrt{3}& \frac{1}{\sqrt{3}}\\ 0& \frac{1}{\sqrt{2}}\end{array}\right]$$

  最后，令
$$S=\left[\begin{array}{cc}1& O^{\text{T}}\\ O& H_2\end{array}\right]H_1=\left[\begin{array}{ccc}1& 0& 0\\ 0& \frac{6-x^2}{x^2+6}& \frac{-2\sqrt{6}x}{x^2+6}\\ 0& \frac{-2\sqrt{6}x}{x^2+6}& \frac{x^2-6}{x^2+6}\end{array}\right]\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ \frac{2}{\sqrt{6}}& \frac{\sqrt{6}-2}{\sqrt{6}-6}& \frac{2}{\sqrt{6}-6}\\ \frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}-6}& \frac{\sqrt{6}-5}{\sqrt{6}-6}\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}& \frac{1}{\sqrt{6}}\\ \frac{1}{\sqrt{3}}& -\frac{1}{\sqrt{3}}& \frac{1}{\sqrt{3}}\\ \frac{1}{\sqrt{2}}& 0& -\frac{1}{\sqrt{2}}\end{array}\right]$$

$$Q=S^{\mathrm{T}}==\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{2}}\\ \frac{2}{\sqrt{6}}& -\frac{1}{\sqrt{3}}& 0\\ \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}& -\frac{1}{\sqrt{2}}\end{array}\right],R=SA=\left[\begin{array}{ccc}\sqrt{6}& \sqrt{6}& \frac{7}{\sqrt{6}}\\ 0& \sqrt{3}& \frac{1}{\sqrt{3}}\\ 0& 0& \frac{1}{\sqrt{2}}\end{array}\right]$$

则有$A=QR$。（感兴趣的读者可用MATLAB验证结果，没想到一个小小的矩阵计算这么复杂(╯﹏╰)）

四、矩阵的满秩分解

  设$A\in C_r^{m\times n}(r>0)$，存在矩阵$F\in C_r^{m\times r}$和$G\in C_r^{r\times n}$，使得
$$A=FG\text{\hspace{1em}(7)}$$

其中$r$为矩阵的秩，$F$是列满秩矩阵，$G$是行满秩矩阵，上式即为矩阵的满秩分解。当$A$是满秩（列满秩或行满秩）矩阵时，$A$可分解为一个因子是单位矩阵，另一个因子是$A$本身，称此满秩分解为平凡分解。

五、矩阵的奇异值分解

  设$A\in C_r^{m\times n}(r>0)$，则存在$m$阶酉矩阵$U$和$n$阶酉矩阵$V$，使得
$$A=UD{V}^{\mathrm{H}}=U\left[\begin{array}{cc}\Sigma & O\\ O& O\end{array}\right]V^{\mathrm{H}}\text{\hspace{1em}(8)}$$

其中，$V^{\mathrm{H}}$代表酉矩阵$V$的共轭转置，$\Sigma =\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}({\sigma }_1,{\sigma }_2,\cdots ,{\sigma }_r)$，${\sigma }_i(i=1,2,\cdots ,r)$为矩阵$A$的全部非零奇异值。上式即为矩阵的奇异值分解（Singular Value Decomposition，简称SVD分解），当$A$为实对称矩阵时，也称正交对角分解。

  矩阵的奇异值分解在最优化问题、特征值问题、最小二乘问题、广义逆矩阵问题及统计学等方面都有重要应用。下面通过一个具体实例，将矩阵$A=\left[\begin{array}{cc}1& 0\\ 0& 1\\ 1& 1\end{array}\right]$进行SVD分解。

• 首先计算$A^{\mathrm{T}}A=\left[\begin{array}{ccc}1& 0& 1\\ 0& 1& 1\end{array}\right]\left[\begin{array}{cc}1& 0\\ 0& 1\\ 1& 1\end{array}\right]=\left[\begin{array}{cc}2& 1\\ 1& 2\end{array}\right]$

• 对矩阵$A^{\mathrm{T}}A$，由$|\lambda I-A^{\mathrm{T}}A|=\left|\begin{array}{cc}\lambda -2& -1\\ -1& \lambda -2\end{array}\right|=(\lambda -1)(\lambda -3)=0$，得特征值${\lambda }_1=1,{\lambda }_2=3$，对应的特征向量分别为：

$${\mathbf{\xi }}_1=k_1\left[\begin{array}{c}1\\ -1\end{array}\right],{\mathbf{\xi }}_1=k_2\left[\begin{array}{c}1\\ 1\end{array}\right]$$

• 于是有

$$\begin{array}{l}\Sigma =\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}(\sqrt{{\lambda }_1},\sqrt{{\lambda }_2})=\left[\begin{array}{cc}1& 0\\ 0& \sqrt{3}\end{array}\right]\\ D=\left[\begin{array}{cc}\Sigma & O\\ O& O\end{array}\right]=\left[\begin{array}{cc}1& 0\\ 0& \sqrt{3}\\ 0& 0\end{array}\right]\\ V=\left[\begin{array}{cc}\frac{{\mathbf{\xi }}_1}{\Vert {\mathbf{\xi }}_1\Vert }& \frac{{\mathbf{\xi }}_2}{\Vert {\mathbf{\xi }}_2\Vert }\end{array}\right]=\left[\begin{array}{cc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\\ -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\end{array}\right],\text{这里V可以有四种解}\end{array}$$

• 计算

$$U_1=AV{\Sigma }^{-1}=\left[\begin{array}{cc}1& 0\\ 0& 1\\ 1& 1\end{array}\right]\left[\begin{array}{cc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\\ -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\end{array}\right]\left[\begin{array}{cc}1& 0\\ 0& \frac{1}{\sqrt{3}}\end{array}\right]=\left[\begin{array}{cc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{6}}\\ -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{6}}\\ 0& \frac{2}{\sqrt{6}}\end{array}\right]$$

• 构造一个列向量$U_2=[{\alpha }_1,{\alpha }_2,{\alpha }_3{]}^{\mathrm{T}}$，使得$U=\left[\begin{array}{cc}{U}_1& U_2\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{6}}& {\alpha }_1\\ -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{6}}& {\alpha }_2\\ 0& \frac{2}{\sqrt{6}}& {\alpha }_3\end{array}\right]$为酉矩阵，即有：

$$\left[\begin{array}{ccc}\frac{1}{\sqrt{2}}& -\frac{1}{\sqrt{2}}& 0\\ \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{6}}& \frac{2}{\sqrt{6}}\end{array}\right]\left[\begin{array}{c}{\alpha }_1\\ {\alpha }_2\\ {\alpha }_3\end{array}\right]=0,且\sqrt{{\alpha }_1^2+{\alpha }_2^2+{\alpha }_3^2}=1$$

得$$U_2=\pm \left[\begin{array}{ccc}\frac{1}{\sqrt{3}}& \frac{1}{\sqrt{3}}& -\frac{1}{\sqrt{3}}\end{array}\right]$$

取$$U=\left[\begin{array}{ccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}\\ -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}\\ 0& \frac{2}{\sqrt{6}}& -\frac{1}{\sqrt{3}}\end{array}\right],\text{这里U可以有两种解}$$

• 最后，$A$的SVD分解为：

$$A=UD{V}^{\mathrm{T}}=\left[\begin{array}{ccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}\\ -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}\\ 0& \frac{2}{\sqrt{6}}& -\frac{1}{\sqrt{3}}\end{array}\right]\left[\begin{array}{cc}1& 0\\ 0& \sqrt{3}\\ 0& 0\end{array}\right]{\left[\begin{array}{cc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\\ -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\end{array}\right]}^{\mathrm{T}}=\left[\begin{array}{cc}1& 0\\ 0& 1\\ 1& 1\end{array}\right]$$

当调整奇异值的顺序，使$\Sigma =\left[\begin{array}{cc}\sqrt{3}& 0\\ 0& 1\end{array}\right]$时，可得
$$\begin{array}{l}D=\left[\begin{array}{cc}\sqrt{3}& 0\\ 0& 1\\ 0& 0\end{array}\right]\\ V=\left[\begin{array}{cc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\\ \frac{1}{\sqrt{2}}& -\frac{1}{\sqrt{2}}\end{array}\right],\text{这里V可以有四种解}\\ U=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{3}}\\ \frac{1}{\sqrt{6}}& -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{3}}\\ \frac{2}{\sqrt{6}}& 0& -\frac{1}{\sqrt{3}}\end{array}\right],\text{这里U可以有两种解}\end{array}$$

$$A=UD{V}^{\mathrm{T}}=\left[\begin{array}{ccc}\frac{1}{\sqrt{6}}& \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{3}}\\ \frac{1}{\sqrt{6}}& -\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{3}}\\ \frac{2}{\sqrt{6}}& 0& -\frac{1}{\sqrt{3}}\end{array}\right]\left[\begin{array}{cc}\sqrt{3}& 0\\ 0& 1\\ 0& 0\end{array}\right]{\left[\begin{array}{cc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\\ \frac{1}{\sqrt{2}}& -\frac{1}{\sqrt{2}}\end{array}\right]}^{\mathrm{T}}=\left[\begin{array}{cc}1& 0\\ 0& 1\\ 1& 1\end{array}\right]$$

可以看到，矩阵的奇异值分解通常不是唯一的。