矩阵分析笔记

矩阵分析

向量序列

• 向量序列记为 { x ( k ) } : x ( 1 ) , x ( 1 ) , … , x ( k ) , … , 其中 x ( k ) ∈ C n 向量序列记为\{x^{(k)}\}:x^{(1)},x^{(1)},\dots,x^{(k)},\dots,其中x^{(k)}\in\mathbb{C}^n 向量序列记为{x(k)}:x(1),x(1),…,x(k),…,其中x(k)∈Cn
• 设 x ( k ) , x ∈ C n , k = 1 , 2 , … ，向量序列 { x ( k ) } 收敛于向量 x ，则 ∥ x ( k ) − x ∥ → 0 , k → + ∞ 设x^{(k)},x\in\mathbb{C}^n,k=1,2,\dots，向量序列\{x^{(k)}\}收敛于向量x，则\begin{Vmatrix}x^{(k)}-x\end{Vmatrix}\rightarrow0,k\rightarrow+\infty 设x(k),x∈Cn,k=1,2,…，向量序列{x(k)}收敛于向量x，则 ​x(k)−x​ ​→0,k→+∞
  • $记为\underset{k\to +\infty }{\lim }{x}^{(k)}=x或x^{(k)}\to x,k\to +\infty$
  • $依范数收敛$

$\text{Ex1}设x^{(k)}=(1+\frac{1}{2^k},1+\frac{1}{3^k},\dots ,1+\frac{1}{(n+1{)}^k}{)}^T,x=(1,1,\dots ,1{)}^T,试证：\underset{k\to +\infty }{\lim }{x}^{(k)}=x$

• $$\begin{array}{rl}& x^{(k)}-x=\left(\begin{array}{cccc}\frac{1}{2^k}& \frac{1}{3^k}& \dots & \frac{1}{(n+1{)}^k}\end{array}\right)\\ & {\Vert \begin{array}{c}{x}^{(k)}-x\end{array}\Vert }_{\infty }=\underset{2\le i\le n+1}{\max }\frac{1}{i^k}=\frac{1}{2^k}\stackrel{k\to +\infty }{\to }0\\ & \therefore \underset{k\to +\infty }{\lim }{x}^{(k)}=x\end{array}$$

Banach空间

• $赋范线性空间中任意收敛向量序列的极限均属于此赋范线性空间$
• 设 x ( k ) = ( x 1 ( k ) , x 2 ( k ) , … , x n ( k ) ) T , x = ( x 1 , x 2 , … , x n ) T ∈ C n , 则向量序列 { x ( k ) } 收敛于 x 设x^{(k)}=(x_1^{(k)},x_2^{(k)},\dots,x_n^{(k)})^T,x=(x_1,x_2,\dots,x_n)^T\in\mathbb{C}^n,则向量序列\{x^{(k)}\}收敛于x 设x(k)=(x1(k)​,x2(k)​,…,xn(k)​)T,x=(x1​,x2​,…,xn​)T∈Cn,则向量序列{x(k)}收敛于x

矩阵序列

矩阵序列的极限运算性质

• $若\underset{k\to +\infty }{\lim }{A}^{(k)}=A,\underset{k\to +\infty }{\lim }{B}^{(k)}=B.A^{(k)},B^{(k)}\in {\mathbb{C}}^{m\times n}$

  • $\underset{k\to +\infty }{\lim }(a{A}^{(k)}\pm b{B}^{(k)})=aA\pm bBa,b\in \mathbb{C}$
  • $\underset{k\to +\infty }{\lim }(A^{(k)}{B}^{(k)})=AB{A}^{(k)},B^{(k)}\in {\mathbb{C}}^{m\times n}$
  • $若A^{(k)},A均为可逆方阵，则\{(A^{(k)}{)}^{-1}\}也收敛，且\underset{k\to +\infty }{\lim }(A^{(k)}{)}^{-1}=A^{-1}$

• 幂矩阵序列 { A ( k ) } : A 1 , A 2 , … , A k , … , A ∈ C n × n 幂矩阵序列\{A^{(k)}\}:A^1,A^2,\dots,A^k,\dots,A\in\mathbb{C}^{n\times n} 幂矩阵序列{A(k)}:A1,A2,…,Ak,…,A∈Cn×n

  • $\underset{k\to +\infty }{\lim }{A}^k=0\iff \rho (A)<1$
  • $\Vert \begin{array}{c}A\end{array}\Vert <1\Rightarrow \underset{k\to +\infty }{\lim }{A}^k=0$
  • $收敛矩阵A：A^k收敛于零矩阵$

$\text{Ex2}A=\left(\begin{array}{ccc}0.2& 0.5& 0.1\\ 0.1& 0.5& 0.3\\ 0.2& 0.4& 0.2\end{array}\right),{\Vert \begin{array}{c}A\end{array}\Vert }_{\infty }=0.9\Rightarrow \underset{k\to +\infty }{\lim }{A}^k=0$

矩阵级数

定义

• ${\sum }_{k=1}^{\infty }{A}^k=A^{(1)}+A^{(2)}+\cdots +A^{(k)}+\dots ,A^{(k)}为矩阵级数一般项$

• $矩阵级数敛散性$

  • $矩阵级数部分和：S^{(N)}=A^{(1)}+A^{(2)}+\cdots +A^{(N)}$
  • $矩阵级数收敛：\underset{N\to \infty }{\lim }{S}^{(N)}=S存在$

• $$A^k=\left(\begin{array}{ccc}{a}_{11}^{(k)}& \dots & a_{1n}^{(k)}\\ ⋮& \ddots & ⋮\\ a_{m1}^{(k)}& \dots & a_{mn}^{(k)}\end{array}\right)S^{(N)}=\left(\begin{array}{ccc}{\sum }_{k=1}^N{a}_{11}^{(k)}& \dots & {\sum }_{k=1}^N{a}_{1n}^{(k)}\\ ⋮& \ddots & ⋮\\ {\sum }_{k=1}^N{a}_{m1}^{(k)}& \dots & {\sum }_{k=1}^N{a}_{mn}^{(k)}\end{array}\right)$$

性质

• ${\sum }_{k=1}^{\infty }{A}^{(k)}收敛\iff mn个数项级数{\sum }_{k=1}^{\infty }{a}_{ij}^{(k)}收敛$

• ${\sum }_{k=1}^{\infty }{A}^{(k)}收敛\Rightarrow \underset{k\to +\infty }{\lim }{A}^{(k)}=0$

• ${\sum }_{k=1}^{\infty }{A}^{(k)}=S_1,{\sum }_{k=1}^{\infty }{B}^{(k)}=S_2,则{\sum }_{k=1}^{\infty }{A}^{(k)}\pm B^{(k)}=S_1\pm S_2$

• ${\sum }_{k=1}^{\infty }{A}^{(k)}=S,\mu \in \mathbb{C},则{\sum }_{k=1}^{\infty }\mu A^{(k)}=\mu S$

绝对收敛

• $若矩阵级数{\sum }_{k=1}^{\infty }{A}^{(k)}绝对收敛：对应的m\times n个数项级数{\sum }_{k=1}^{\infty }{a}_{ij}^{(k)}绝对收敛$

• $定理：{\sum }_{k=1}^{\infty }{A}^{(k)}绝对收敛\iff 对\forall 范数\Vert \begin{array}{c}\cdot \end{array}\Vert ,{\sum }_{k=1}^{\infty }\Vert \begin{array}{c}{A}^{(k)}\end{array}\Vert 收敛$

矩阵幂级数

• ${\sum }_{k=0}^{\infty }{c}_k{A}^k=c_{0}I+c_{1}A+c_2{A}^2+\cdots +c_k{A}^k+\dots$

• $定理：正项级数|c_0|\Vert \begin{array}{c}I\end{array}\Vert +{\sum }_{k=1}^{\infty }|c_k|{\Vert \begin{array}{c}A\end{array}\Vert }^k收敛，则{\sum }_{k=0}^{\infty }{c}_k{A}^k绝对收敛$

  • $若矩阵A在某一种范数\Vert \begin{array}{c}A\end{array}\Vert 在纯量幂级数{\sum }_{k=0}^{\infty }{c}_k{z}^k的收敛圆内，则矩阵{\sum }_{k=0}^{\infty }{c}_k{A}^k收敛$

$\text{Ex3}A=\left(\begin{array}{ccc}0.2& 0.5& 0.1\\ 0.1& 0.5& 0.3\\ 0.2& 0.4& 0.2\end{array}\right),则I+A+A^2+\dots 绝对收敛$

• $定理$

  • $设A^{(k)}\in {\mathbb{C}}^{n\times n},如果A的谱半径\rho (A)的值在纯量幂级数{\sum }_{k=0}^{\infty }{c}_k{z}^k的收敛圆内，则{\sum }_{k=0}^{\infty }{c}_k{A}^k绝对收敛$

  • $如果A的特征值中有一个在{\sum }_{k=0}^{\infty }{c}_k{z}^k收敛圆外，则{\sum }_{k=0}^{\infty }{c}_k{A}^k发散$

  • $定理条件要求\rho (A)<R$

    • $R=\frac{1}{\rho },\rho =\underset{k\to \infty }{\lim }\frac{|c_{k+1}|}{|c_k|}$

• $$定理：\begin{array}{rl}& 矩阵幂级数\sum _{k=0}^{\infty }{A}^k=I+A+A^2+\cdots +A^k+\dots 绝对收敛\\ & \iff \rho (A)<1,且I+A+A^2+\cdots +A^k+\cdots =(I-A{)}^{-1}\end{array}$$

$\text{Ex4}设A=\left(\begin{array}{cc}0.1& 0.7\\ 0.3& 0.6\end{array}\right),求{\sum }_{k=0}^{\infty }{A}^k$

• $$\begin{array}{rl}& |\lambda I-A|=\left|\begin{array}{cc}\lambda -1& -0.7\\ -0.3& \lambda -0.6\end{array}\right|={\lambda }^2-0.7\lambda -0.15=0\\ & \Rightarrow \lambda =\frac{0.7\pm \sqrt{1.09}}{2}\Rightarrow \rho (A)=\frac{0.7+\sqrt{1.09}}{2}<1\Rightarrow \sum _{k=0}^{\infty }{A}^k绝对收敛\\ & 则\sum _{k=0}^{\infty }{A}^k=(1-A{)}^{-1}=\frac{1}{0.15}\left(\begin{array}{cc}0.4& 0.7\\ 0.3& 0.9\end{array}\right)\end{array}$$

• $矩阵幂级数{\sum }_{k=0}^{\infty }{c}_k{A}^k的敛散性对应纯量幂级数{\sum }_{k=0}^{\infty }{c}_k{z}^k的敛散性$

• $纯量幂级数的和函数S(z)={\sum }_{k=0}^{\infty }{c}_k{z}^k在其收敛域(-R,R)内连续、可积、可微$

  $对应矩阵幂级数的和函数f(A)={\sum }_{k=0}^{\infty }{c}_k{A}^k,\rho (A)<R$

• $在\rho (A<R时，利用前k+1项和求，和函数的近似值：$

  $f(A)={\sum }_{k=0}^{\infty }{c}_k{A}^k\approx c_{0}I+c_{1}A+c_2{A}^2+\cdots +c_k{A}^k误差较小$

• $定理：\Vert \begin{array}{c}A\end{array}\Vert <1,\forall 非负整数k，有\Vert \begin{array}{c}(1-A{)}^{-1}-(I+A+A^2+\cdots +A^k)\end{array}\Vert <\frac{{\Vert \begin{array}{c}A\end{array}\Vert }^{k+1}}{1-\Vert \begin{array}{c}A\end{array}\Vert }$

$\text{Ex5}设A=\left(\begin{array}{ccc}-1& 1& 1\\ -5& 21& 17\\ 6& -26& -21\end{array}\right),讨论{\sum }_{k=0}^{\infty }(-1{)}^k\frac{1}{3^{k+1}{k}^2}{A}^k的敛散性$

• $$\begin{array}{rl}& 求收敛半径R：c_k=(-1{)}^k\frac{1}{3^{k+1}{k}^2}\Rightarrow \rho =\underset{k\to \infty }{\lim }\frac{|c_{k+1}|}{|c_k|}=\frac{1}{3}\Rightarrow R=\frac{1}{\rho }=3\\ & 求谱半径\rho (A)：|\lambda I-A|=\left|\begin{array}{ccc}\lambda +1& -1& -1\\ 5& \lambda -21& -17\\ -6& 26& \lambda +21\end{array}\right|={\lambda }^2(\lambda +1)=0\Rightarrow \rho (A)=1\\ & 判断敛散性:\rho (A)=1<R,收敛\end{array}$$

$\text{Ex6}设A=\left(\begin{array}{ccc}0& a& a\\ a& 0& a\\ a& a& 0\end{array}\right),求a为何值时，{\sum }_{k=0}^{\infty }(-1{)}^{k-1}\frac{1}{2^{k}k}{A}^k收敛$
$$\begin{array}{rl}& c_k=(-1{)}^{k-1}\frac{1}{2^{k}k}\Rightarrow R=2\\ & \dots \end{array}$$

矩阵函数

常见矩阵函数

• $一般形式：\forall A\in {\mathbb{C}}^{n\times n}\to f(A)={\sum }_{k=0}^{\infty }{c}_k{A}^k$

  • $纯量幂级数f(z)={\sum }_{k=0}^{\infty }{c}_k{z}^k\Rightarrow 矩阵幂级数f(A)={\sum }_{k=0}^{\infty }{c}_k{A}^k$
  • $矩阵指数函数：e^A={\sum }_{k=0}^{\infty }\frac{A^k}{k!}$
  • $矩阵三角函数：\sin A={\sum }_{k=0}^{\infty }(-1{)}^k\frac{A^{2k+1}}{(2k+1)!}，\cos A={\sum }_{k=0}^{\infty }(-1{)}^k\frac{A^{2k}}{(2k)!}$

• $定理：AB=BA\to e^A\cdot e^B=e^B\cdot e^A=e^{A+B}$

  • $\forall A\in {\mathbb{C}}^{n\times n},e^A总可逆，且(e^A{)}^{-1}=e^{-A}$
    $$\begin{array}{rl}& \because e^A\cdot e^{-A}=e^{A+(-A)}=e^0,e^A=\sum _{k=0}^{\infty }\frac{A^k}{k!}\Rightarrow e^0=\sum _{k=0}^{\infty }\frac{0^k}{k!}=I\\ & \therefore e^A\cdot e^{-A}=I\Rightarrow (e^A{)}^{-1}=e^{-A}\end{array}$$

  • $\forall A\in {\mathbb{C}}^{n\times n},有(e^A{)}^m=e^{mA}$

矩阵函数值的计算

$凯莱-哈密顿定理$

$利用相似矩阵$

• $A可对角化：f(A)=C\text{diag}\left(\begin{array}{cccc}f({\lambda }_1)& f({\lambda }_2)& \dots & f({\lambda }_n)\end{array}\right)C^{-1}$

$\text{Ex7}设A=\left(\begin{array}{cc}3& 2\\ 1& 2\end{array}\right),求A^4+2A^3+A及e^A,\sin A$

• $$\begin{array}{rl}& (1)求A的特征值、特征向量\\ & |\lambda I-A|=\left(\begin{array}{cc}\lambda -3& -2\\ -1& \lambda -2\end{array}\right)={\lambda }^2-5\lambda +4\Rightarrow {\lambda }_1=1,{\lambda }_2=4\\ & 分别对应特征向量{\left(\begin{array}{cc}-1& 1\end{array}\right)}^T,{\left(\begin{array}{cc}2& 1\end{array}\right)}^T\\ & (2)确定相似变换和对角阵\\ & C=\left(\begin{array}{cc}-1& 2\\ 1& 1\end{array}\right),C^{-1}=\left(\begin{array}{cc}-\frac{1}{3}& \frac{2}{3}\\ \frac{1}{3}& \frac{1}{3}\end{array}\right)\Rightarrow C^{-1}AC=\Lambda =\left(\begin{array}{cc}1& 0\\ 0& 4\end{array}\right)\\ & \Rightarrow A=C\left(\begin{array}{cc}1& 0\\ 0& 4\end{array}\right)C^{-1}\\ & (3)计算函数值\\ & f(A)=C\text{diag}\left(\begin{array}{cc}f({\lambda }_1)& f({\lambda }_2)\end{array}\right)C^{-1}\\ & A^4+2A^3+A=\left(\begin{array}{cc}-1& 2\\ 1& 1\end{array}\right)\left(\begin{array}{cc}4& 0\\ 0& 388\end{array}\right)\left(\begin{array}{cc}-\frac{1}{3}& \frac{2}{3}\\ \frac{1}{3}& \frac{1}{3}\end{array}\right)=\left(\begin{array}{cc}260& 256\\ 128& 132\end{array}\right)\\ & e^A=\left(\begin{array}{cc}-1& 2\\ 1& 1\end{array}\right)\left(\begin{array}{cc}{e}^1& 0\\ 0& e^4\end{array}\right)\left(\begin{array}{cc}-\frac{1}{3}& \frac{2}{3}\\ \frac{1}{3}& \frac{1}{3}\end{array}\right)=\frac{1}{3}\left(\begin{array}{cc}{e}^1+2e^4& -2e^1+e^4\\ -e^1+e^4& 2e^1+e^4\end{array}\right)\\ & \sin A=\left(\begin{array}{cc}-1& 2\\ 1& 1\end{array}\right)\left(\begin{array}{cc}\sin 1& 0\\ 0& \sin 4\end{array}\right)\left(\begin{array}{cc}-\frac{1}{3}& \frac{2}{3}\\ \frac{1}{3}& \frac{1}{3}\end{array}\right)=\frac{1}{3}\left(\begin{array}{cc}\sin 1+2\sin 4& -2\sin 1+\sin 4\\ -\sin 1+\sin 4& 2\sin 1+\sin 4\end{array}\right)\end{array}$$

• $A不可对角化：A\sim J,即\exists C,C^{-1}AC=J\Rightarrow A=CJ{C}^{-1}$

  $定义矩阵函数f(A)=Cf(J)C^{-1}=C\left(\begin{array}{cccc}f(J_1({\lambda }_1))& & & \\ & f(J_2({\lambda }_2)& & \\ & & \ddots & \\ & & & f(J_r({\lambda }_r))\end{array}\right)C^{-1}(\ast )$

  $其中f_i(J_i({\lambda }_i))=\left(\begin{array}{ccccc}f({\lambda }_i)& f^{\prime }({\lambda }_i)& \frac{1}{2}{f}^{\prime \prime }({\lambda }_i)& \dots & \frac{1}{(n_i-1)!}{f}^{(n_i-1)}({\lambda }_i)\\ & f({\lambda }_i)& f^{\prime }({\lambda }_i)& \dots & \frac{1}{(n_i-1)!}{f}^{(n_i-1)}({\lambda }_i)\\ & & \ddots & \ddots & ⋮\\ & & & \ddots & f^{\prime }({\lambda }_i)\\ & & & & f({\lambda }_i)\end{array}\right),(\ast )为f(A)的若当标准型$

  • $对于三阶方阵A，有f(A)=Cf(J)C^{-1},J=\left(\begin{array}{ccc}{J}_1& & \\ & \ddots & \\ & & J_r\end{array}\right)$

• $$\begin{array}{rl}& J_i=({\lambda }_i),r=3\Rightarrow J=\Lambda =\left(\begin{array}{ccc}{\lambda }_1& & \\ & {\lambda }_2& \\ & & {\lambda }_3\end{array}\right)\\ & \Rightarrow f(A)=Cf(\Lambda )C^{-1}=C\text{diag}\left(\begin{array}{cccc}f({\lambda }_1)& f({\lambda }_2)& \dots & f({\lambda }_n)\end{array}\right)C^{-1}\\ & J_1=({\lambda }_1),J_2=\left(\begin{array}{cc}{\lambda }_2& 1\\ 0& {\lambda }_2\end{array}\right)\Rightarrow J=\left(\begin{array}{cc}{J}_1& 0\\ 0& J_2\end{array}\right)\\ & \Rightarrow f(J_1)=f({\lambda }_1),f(J_2)=\left(\begin{array}{cc}f({\lambda }_2)& f^{\prime }({\lambda }_2)\\ 0& f({\lambda }_2)\end{array}\right)\\ & J_1=\left(\begin{array}{ccc}{\lambda }_1& 1& 0\\ 0& {\lambda }_2& 1\\ 0& 0& {\lambda }_3\end{array}\right)\Rightarrow f(J_1)=\left(\begin{array}{ccc}f({\lambda }_1)& f^{\prime }({\lambda }_1)& \frac{1}{2}{f}^{\prime \prime }({\lambda }_1)\\ & f({\lambda }_1)& f^{\prime }({\lambda }_1)\\ & & f({\lambda }_1)\end{array}\right)\end{array}$$

$\text{Ex8}设A=\left(\begin{array}{ccc}3& 1& -1\\ 1& 2& -1\\ 2& 1& 0\end{array}\right),求e^{At}及\frac{1}{A}$

• $$\begin{array}{rl}& |\lambda I-A|=(\lambda -1)(\lambda -2{)}^2\Rightarrow {\lambda }_1=1,{\lambda }_{2,3}=2\\ & 令相似变换P=\left(\begin{array}{ccc}\overrightarrow{x_1}& \overrightarrow{x_2}& \overrightarrow{x_3}\end{array}\right)，则AP=PJ，得A\left(\begin{array}{ccc}\overrightarrow{x_1}& \overrightarrow{x_2}& \overrightarrow{x_3}\end{array}\right)=\left(\begin{array}{ccc}\overrightarrow{x_1}& \overrightarrow{x_2}& \overrightarrow{x_3}\end{array}\right)\left(\begin{array}{ccc}1& & \\ & 2& 1\\ & & 2\end{array}\right)\\ & {\lambda }_1=1\Rightarrow \overrightarrow{x_1}={\left(\begin{array}{ccc}0& 1& 1\end{array}\right)}^T\\ & {\lambda }_2=2\Rightarrow \overrightarrow{x_2}={\left(\begin{array}{ccc}1& 0& 1\end{array}\right)}^T\\ & {\lambda }_3=2\Rightarrow 求解(2I-A)\overrightarrow{x_3}=-\overrightarrow{x_2},得广义特征向量\overrightarrow{x_3}={\left(\begin{array}{ccc}1& 1& 1\end{array}\right)}^T\\ & P=\left(\begin{array}{ccc}0& 1& 1\\ 1& 0& 1\\ 1& 1& 1\end{array}\right)\Rightarrow P^{-1}=\left(\begin{array}{ccc}-1& 0& 1\\ 0& -1& 1\\ 1& 1& -1\end{array}\right)\Rightarrow A=PJ{P}^{-1}=P\left(\begin{array}{ccc}1& & \\ & 2& 1\\ & & 2\end{array}\right)P^{-1}\\ & f(A)=P\left(\begin{array}{cc}f(J_1({\lambda }_1))& \\ & f(J_2({\lambda }_2))\end{array}\right)P^{-1}\\ & 求e^{At}:f(J_1)=f(1)=e^t,f(J_2)=\left(\begin{array}{cc}{e}^{2t}& t{e}^{2t}\\ & e^{2t}\end{array}\right)\Rightarrow e^{At}=P\left(\begin{array}{ccc}{e}^t& & \\ & e^{2t}& t{e}^{2t}\\ & & e^{2t}\end{array}\right)P^{-1}\\ & 求\frac{1}{A}:f(J_1)=f(1)=1,f({\lambda }_2)=\frac{1}{{\lambda }_2},f^{\prime }({\lambda }_2)=-\frac{1}{{\lambda }_2^2}\Rightarrow f(J_2)=\left(\begin{array}{cc}\frac{1}{2}& -\frac{1}{4}\\ 0& \frac{1}{2}\end{array}\right)\\ & \dots \end{array}$$