矩阵论
1. 准备知识——复数域上矩阵,Hermite变换)
1.准备知识——复数域上的内积域正交阵
1.准备知识——Hermite阵,二次型,矩阵合同,正定阵,幂0阵,幂等阵,矩阵的秩
2. 矩阵分解——SVD准备知识——奇异值
2. 矩阵分解——SVD
2. 矩阵分解——QR分解
2. 矩阵分解——正定阵分解
2. 矩阵分解——单阵谱分解
2. 矩阵分解——正规分解——正规阵
2. 矩阵分解——正规谱分解
2. 矩阵分解——高低分解
3. 矩阵函数——常见解析函数
3. 矩阵函数——谱公式,幂0与泰勒计算矩阵函数
3. 矩阵函数——矩阵函数求导
4. 矩阵运算——观察法求矩阵特征值特征向量
4. 矩阵运算——张量积
4. 矩阵运算——矩阵拉直
4.矩阵运算——广义逆——加号逆定义性质与特殊矩阵的加号逆
4. 矩阵运算——广义逆——加号逆的计算
4. 矩阵运算——广义逆——加号逆应用
4. 矩阵运算——广义逆——减号逆
5. 线性空间与线性变换——线性空间
5. 线性空间与线性变换——生成子空间
5. 线性空间与线性变换——线性映射与自然基分解,线性变换
6. 正规方程与矩阵方程求解
7. 范数理论——基本概念——向量范数与矩阵范数
7.范数理论——基本概念——矩阵范数生成向量范数&谱范不等式
7. 矩阵理论——算子范数
7.范数理论——范数估计——许尔估计&谱估计
7. 范数理论——非负/正矩阵
8. 常用矩阵总结——秩1矩阵,优阵(单位正交阵),Hermite阵
8. 常用矩阵总结——镜面阵,正定阵
8. 常用矩阵总结——单阵,正规阵,幂0阵,幂等阵,循环阵
A = ( a 1 b 1 a 1 b 2 ⋯ a 1 b n a 2 b 1 a 2 b 2 ⋯ a 2 b n ⋮ ⋮ ⋱ ⋮ a n b 1 a n b 2 ⋯ a n b n ) n × n = ( a 1 a 2 ⋮ a n ) ( b 1 b 2 ⋯ b n ) = Δ α β T 其中 α = ( a 1 a 2 ⋮ a n ) , β = ( b 1 b 2 ⋮ b n ) \begin{aligned} A&=\left( \begin{matrix} a_1b_1\quad &a_1b_2\quad &\cdots\quad &a_1b_n\\ a_2b_1\quad &a_2b_2\quad &\cdots\quad &a_2b_n\\ \vdots\quad &\vdots\quad &\ddots\quad &\vdots\\ a_nb_1\quad &a_nb_2\quad &\cdots\quad &a_nb_n \end{matrix} \right)_{n\times n}=\left( \begin{matrix} a_1\\a_2\\\vdots \\a_n \end{matrix} \right)\left( b_1\quad b_2\quad \cdots \quad b_n \right)\overset{\Delta}{=}\alpha \beta^{T}\\\\ &其中 \alpha=\left( \begin{matrix} a_1\\a_2\\\vdots \\a_n \end{matrix} \right),\beta=\left( \begin{matrix} b_1\\b_2\\\vdots \\b_n \end{matrix} \right) \end{aligned} A= a1b1a2b1⋮anb1a1b2a2b2⋮anb2⋯⋯⋱⋯a1bna2bn⋮anbn n×n= a1a2⋮an (b1b2⋯bn)=ΔαβT其中α= a1a2⋮an ,β= b1b2⋮bn
∣ λ I n − A ∣ = ∣ λ I n − α n × 1 β 1 × n T ∣ = 换位公式 : ∣ λ n − ( A B ) n ∣ = λ n − p ∣ λ I p − ( B A ) p ∣ λ n − 1 ∣ λ I 1 − β 1 × n T α n × 1 I 1 ∣ = λ n − 1 ( λ I − t r ( A ) ) , 其中 t r ( A ) = ∑ i = 1 n a i i b i i \begin{aligned} &\mid \lambda I_n-A\mid = \mid \lambda I_n-\alpha_{n\times 1}\beta_{1\times n}^T\mid\xlongequal{换位公式:\vert\lambda_n-(AB)_n \vert=\lambda^{n-p}\vert \lambda I_p-(BA)_p\vert}\lambda^{n-1}\mid\lambda I_1-\beta_{1\times n}^T\alpha_{n\times 1}I_1\mid\\ &=\lambda^{n-1}(\lambda I-tr(A)),其中 tr(A)=\sum\limits_{i=1}\limits^{n}a_{ii}b_{ii} \end{aligned} ∣λIn−A∣=∣λIn−αn×1β1×nT∣换位公式:∣λn−(AB)n∣=λn−p∣λIp−(BA)p∣λn−1∣λI1−β1×nTαn×1I1∣=λn−1(λI−tr(A)),其中tr(A)=i=1∑naiibii
eg
若 A = A n × n A = A_{n\times n} A=An×n , r ( A ) = 1 r(A)=1 r(A)=1 ,则全体特征值为 λ ( A ) = { t r ( A ) , 0 , . . . , 0 } \lambda(A)=\{tr(A),0,...,0\} λ(A)={tr(A),0,...,0} ,其中 t r ( A ) = a 1 b 1 + a 2 b 2 + . . . + a n b n = β T α tr(A)=a_1b_1+a_2b_2+...+a_nb_n=\beta^T\alpha tr(A)=a1b1+a2b2+...+anbn=βTα
证明:
由换位公式可知, α n × 1 β 1 × n T \alpha_{n\times 1}\beta_{1\times n}^T αn×1β1×nT 与 β 1 × n T α n × 1 \beta_{1\times n}^T\alpha_{n\times 1} β1×nTαn×1 相差 n-1 个零根,即有一个相等的非零特征根,而 β 1 × n T α n × 1 \beta_{1\times n}^T\alpha_{n\times 1} β1×nTαn×1 为1阶矩阵,所以 λ 1 = β 1 × n T α n × 1 = a 1 b 1 + a 2 b 2 + . . . + a n b n = t r ( A ) \lambda_1=\beta_{1\times n}^T\alpha_{n\times 1}=a_1b_1+a_2b_2+...+a_nb_n=tr(A) λ1=β1×nTαn×1=a1b1+a2b2+...+anbn=tr(A)
A = α β T A=\alpha \beta^T A=αβT 的列向量都是 λ 1 = t r ( A ) \lambda_1=tr(A) λ1=tr(A) 的特征向量
证明:
A α = ( α β ) α = λ 1 α \begin{aligned} A\alpha = (\alpha \beta)\alpha=\lambda_1 \alpha \end{aligned} Aα=(αβ)α=λ1α
A 为秩 1 矩阵, λ ( A ) = { t r ( A ) , 0 , 0 } = { − 2 , 0 , 0 } 可知 ∣ λ I − A ∣ = x 2 ( x + 2 ) , 其中 λ 1 = − 2 , 可取 ( 1 1 2 ) 为 A 的特向, A ( 1 1 2 ) = − 2 ( 1 1 2 ) \begin{aligned} &A为秩1矩阵,\lambda(A)=\{tr(A),0,0\}=\{-2,0,0\}\\ &可知 \vert \lambda I-A\vert=x^2(x+2),其中\lambda_1=-2,可取\left(\begin{matrix}1\\1\\2\end{matrix}\right)为A的特向,A\left(\begin{matrix}1\\1\\2\end{matrix}\right)=-2\left(\begin{matrix}1\\1\\2\end{matrix}\right) \end{aligned} A为秩1矩阵,λ(A)={tr(A),0,0}={−2,0,0}可知∣λI−A∣=x2(x+2),其中λ1=−2,可取 112 为A的特向,A 112 =−2 112
A 为秩 1 矩阵,全体特根 λ ( A ) = { t r ( A ) , 0 , 0 } = { 9 , 0 , 0 } , 可知 ∣ λ I − A ∣ = λ 2 ( λ − 9 ) λ 1 = 9 , 可知 A 的列向量 ( 1 1 − 1 ) 为一个特向, A ( 1 1 − 1 ) = 9 ( 1 1 − 1 ) \begin{aligned} &A为秩1矩阵,全体特根\lambda(A)=\{tr(A),0,0\}=\{9,0,0\},可知 \vert \lambda I-A\vert=\lambda^2(\lambda-9)\\ &\lambda_1=9,可知A的列向量\left(\begin{matrix}1\\1\\-1\end{matrix}\right)为一个特向,A\left(\begin{matrix}1\\1\\-1\end{matrix}\right)=9\left(\begin{matrix}1\\1\\-1\end{matrix}\right) \end{aligned} A为秩1矩阵,全体特根λ(A)={tr(A),0,0}={9,0,0},可知∣λI−A∣=λ2(λ−9)λ1=9,可知A的列向量 11−1 为一个特向,A 11−1 =9 11−1
预:非单位列向量
半: p p p 个 n n n 维列向量 ( p < n ) (p
α 1 , α 2 , ⋯ , α p \alpha_1,\alpha_2,\cdots,\alpha_p α1,α2,⋯,αp 是 n n n 维列向量,且 p ≤ n p\le n p≤n ,且 α 1 ⊥ α 2 ⊥ ⋯ ⊥ α p \alpha_1\bot\alpha_2\bot \cdots\bot\alpha_p α1⊥α2⊥⋯⊥αp
,则称 A = ( α 1 , α 2 , ⋯ , α p ) A=(\alpha_1,\alpha_2,\cdots,\alpha_p) A=(α1,α2,⋯,αp) 为预半优阵
判定
A = ( α 1 , α 2 , ⋯ , α p ) A=(\alpha_1,\alpha_2,\cdots,\alpha_p) A=(α1,α2,⋯,αp) 是预半优阵 ⟺ A H A = ( ( α 1 , α 1 ) ⋯ 0 ⋮ ⋱ 0 0 ⋯ ( α p , α p ) ) \iff A^HA=\left( \begin{matrix} (\alpha_1,\alpha_1)&\cdots&0\\ \vdots&\ddots&0\\ 0&\cdots&(\alpha_p,\alpha_p) \end{matrix} \right) ⟺AHA= (α1,α1)⋮0⋯⋱⋯00(αp,αp) 是对角阵,其中 α 1 , α 2 , ⋯ , α p \alpha_1,\alpha_2,\cdots,\alpha_p α1,α2,⋯,αp 是 n n n 维列向量
区分: A H A A^HA AHA 是 p × p p \times p p×p 阶满秩方阵,而 A A H AA^H AAH 是 n × n n\times n n×n 不满秩方阵
A = ( α 1 , α 2 , ⋯ , α p ) A=(\alpha_1,\alpha_2,\cdots,\alpha_p) A=(α1,α2,⋯,αp) 是预半优阵,其中 α i \alpha_i αi 是 n n n 维列向量,若满足 ∣ α 1 ∣ = ∣ α 2 ∣ = ⋯ = ∣ α p ∣ = 1 \vert \alpha_1 \vert=\vert \alpha_2 \vert=\cdots=\vert \alpha_p \vert = 1 ∣α1∣=∣α2∣=⋯=∣αp∣=1 ,则A为半优阵
判定
A = ( α 1 , ⋯ , α p ) A=(\alpha_1,\cdots,\alpha_p) A=(α1,⋯,αp) 是半优阵 ⟺ α 1 ⊥ ⋯ ⊥ α p \iff \alpha_1\bot\cdots\bot\alpha_p ⟺α1⊥⋯⊥αp ,且 ∣ α 1 ∣ = ⋯ = ∣ α p ∣ = 1 \vert \alpha_1 \vert=\cdots=\vert \alpha_p \vert=1 ∣α1∣=⋯=∣αp∣=1 ⟺ A H A = I p \iff A^HA=I_{p} ⟺AHA=Ip
性质
保模长 A为半优阵,则 ∣ A x ∣ 2 = ∣ x ∣ 2 \vert Ax \vert^2=\vert x \vert^2 ∣Ax∣2=∣x∣2
∣ A x ∣ 2 = ( A x ) H ( A x ) = x H A H A x = ∣ X ∣ 2 \vert Ax \vert^2=(Ax)^H(Ax)=x^HA^HAx=\vert X\vert^2 ∣Ax∣2=(Ax)H(Ax)=xHAHAx=∣X∣2
保正交 A为半优阵, x ⊥ y x\bot y x⊥y ,则 A x ⊥ A y Ax\bot Ay Ax⊥Ay
α 1 , α 2 , ⋯ , α n \alpha_1,\alpha_2,\cdots,\alpha_n α1,α2,⋯,αn 是 n n n 维列向量,且 α 1 ⊥ α 2 ⊥ ⋯ ⊥ α n \alpha_1\bot\alpha_2\bot\cdots\bot\alpha_n α1⊥α2⊥⋯⊥αn ,则 A = ( α 1 , α 2 , ⋯ , α n ) A=(\alpha_1,\alpha_2,\cdots,\alpha_n) A=(α1,α2,⋯,αn) 是预优阵
eg
X 1 = ( 1 i i ) , X 2 = ( 2 i 1 1 ) , X 3 = ( 0 1 − 1 ) ( X 1 , X 2 ) = 0 , ( X 2 , X 3 ) = 0 , ( X 1 , X 3 ) = 0 , ∴ X 1 ⊥ X 2 ⊥ X 3 , A = ( X 1 , X 2 , X 3 ) 是预 − 优阵 \begin{aligned} &X_1=\left( \begin{matrix} 1\\i\\i \end{matrix} \right),X_2=\left( \begin{matrix} 2i\\1\\1 \end{matrix} \right),X_3=\left( \begin{matrix} 0\\1\\-1 \end{matrix} \right)\\\\ &(X_1,X_2)=0,(X_2,X_3)=0,(X_1,X_3)=0,\\\\ &\therefore X_1\bot X_2\bot X_3,A=(X_1,X_2,X_3)是预-优阵 \end{aligned} X1= 1ii ,X2= 2i11 ,X3= 01−1 (X1,X2)=0,(X2,X3)=0,(X1,X3)=0,∴X1⊥X2⊥X3,A=(X1,X2,X3)是预−优阵
判定
A = ( α 1 , α 2 , ⋯ , α n ) ⟺ A H A = ( ( α 1 , α 1 ) ⋯ 0 ⋮ ⋱ 0 0 ⋯ ( α n , α n ) ) 是对角阵 其中, α 1 , α 2 , ⋯ , α p 是 n 维列向量 \begin{aligned} A&=(\alpha_1,\alpha_2,\cdots,\alpha_n)\\ &\iff A^HA=\left( \begin{matrix} (\alpha_1,\alpha_1)&\cdots&0\\ \vdots&\ddots&0\\ 0&\cdots&(\alpha_n,\alpha_n) \end{matrix} \right)是对角阵\\ &其中,\alpha_1,\alpha_2,\cdots,\alpha_p是n维列向量 \end{aligned} A=(α1,α2,⋯,αn)⟺AHA= (α1,α1)⋮0⋯⋱⋯00(αn,αn) 是对角阵其中,α1,α2,⋯,αp是n维列向量
α 1 , α 2 , ⋯ , α n \alpha_1,\alpha_2,\cdots,\alpha_n α1,α2,⋯,αn 是 n n n 维列向量, α 1 ⊥ α 2 ⊥ ⋯ ⊥ α n \alpha_1\bot\alpha_2\bot\cdots\bot\alpha_n α1⊥α2⊥⋯⊥αn 且 ∣ α 1 ∣ = ⋯ = ∣ α n ∣ = 1 \vert \alpha_1 \vert=\cdots=\vert \alpha_n \vert=1 ∣α1∣=⋯=∣αn∣=1 ,则 A A A 是一个优阵(正交阵)
A = ( α 1 , ⋯ , α n ) ⟺ A H A = ( 1 0 ⋯ 0 0 1 ⋯ 0 ⋮ ⋮ ⋱ ⋮ 0 0 ⋯ 1 ) = I 是单位阵 \begin{aligned} A&=(\alpha_1,\cdots,\alpha_n)\\ &\iff A^HA=\left( \begin{matrix} 1&0&\cdots&0\\ 0&1&\cdots&0\\ \vdots&\vdots&\ddots&\vdots\\ 0&0&\cdots&1 \end{matrix} \right)=I是单位阵 \end{aligned} A=(α1,⋯,αn)⟺AHA= 10⋮001⋮0⋯⋯⋱⋯00⋮1 =I是单位阵
A A A 是优阵 ⟺ A H A = I ⟺ A − 1 A = I ⟺ A A H = I \iff A^HA=I\iff A^{-1}A=I\iff AA^H=I ⟺AHA=I⟺A−1A=I⟺AAH=I ⟺ A = ( α 1 , α 2 , ⋯ , α n ) \iff A=(\alpha_1,\alpha_2,\cdots,\alpha_n) ⟺A=(α1,α2,⋯,αn) ,且 α 1 ⊥ α 2 , ⋯ ⊥ α n \alpha_1\bot\alpha_2,\cdots\bot\alpha_n α1⊥α2,⋯⊥αn , ∣ α 1 ∣ = ⋯ = ∣ α n ∣ = 1 \vert \alpha_1\vert=\cdots=\vert\alpha_n\vert=1 ∣α1∣=⋯=∣αn∣=1
∣ A x ∣ 2 = ∣ x ∣ 2 \vert Ax\vert^2=\vert x \vert^2 ∣Ax∣2=∣x∣2 , A A A 是优阵
∵ ∣ A x ∣ 2 = ( A x ) H ( A x ) = x H A H A x = x H I x = ( x , x ) = ∣ x ∣ 2 \because \vert Ax\vert^2 = (Ax)^H(Ax)=x^HA^HAx=x^HIx=(x,x)=\vert x \vert^2 ∵∣Ax∣2=(Ax)H(Ax)=xHAHAx=xHIx=(x,x)=∣x∣2
x ⊥ y ⇒ A x ⊥ A y x\bot y \Rightarrow Ax\bot Ay x⊥y⇒Ax⊥Ay , A A A 是优阵
∵ ( A x , A y ) = ( A y ) H A x = y H A H A x = ( x , y ) = 0 ⟺ A x ⊥ A y \because (Ax,Ay)=(Ay)^HAx=y^HA^HAx=(x,y)=0\iff Ax\bot Ay ∵(Ax,Ay)=(Ay)HAx=yHAHAx=(x,y)=0⟺Ax⊥Ay
( A x , A y ) = ( x , y ) (Ax,Ay)=(x,y) (Ax,Ay)=(x,y), A A A 是优阵
预优阵到优阵
A = ( α 1 , ⋯ , α n ) A=(\alpha_1,\cdots,\alpha_n) A=(α1,⋯,αn) 是预优阵 ⇒ A = ( α 1 ∣ α 1 ∣ , α 2 ∣ α 2 ∣ , ⋯ , α n ∣ α n ∣ ) \Rightarrow A=(\frac{\alpha_1}{\vert \alpha_1\vert},\frac{\alpha_2}{\vert \alpha_2\vert},\cdots,\frac{\alpha_n}{\vert \alpha_n \vert}) ⇒A=(∣α1∣α1,∣α2∣α2,⋯,∣αn∣αn) 是优阵
优阵到优阵
若 A = ( α 1 , α 2 , ⋯ , α n ) A=(\alpha_1,\alpha_2,\cdots,\alpha_n) A=(α1,α2,⋯,αn) 为优阵,则
向量构造优阵
将向量作为镜面阵的法向量,构造镜面阵(优阵+H阵)
若 α = ( a 1 a 2 ⋮ a n ) ∈ C \alpha=\left( \begin{matrix} a_1\\a_2\\\vdots \\ a_n \end{matrix} \right)\in C α= a1a2⋮an ∈C , A = I n − 2 α α H ∣ α ∣ 2 A=I_n-\frac{2\alpha\alpha^H}{\vert \alpha \vert^2} A=In−∣α∣22ααH 是一个优阵
A H = A A^H=A AH=A 且 A 2 = I ( A − 1 = A ) A^2=I(A^{-1}=A) A2=I(A−1=A)
A A A 为优阵 ( A H A = I ) (A^HA=I) (AHA=I)
1. A 2 = ( I n − 2 α α H ∣ α ∣ 2 ) ( I n − 2 α α H ∣ α ∣ 2 ) = I n 2 − 4 α α H ∣ α ∣ 2 + 4 ( α α H ) ( α α H ) ∣ α ∣ 4 = I n − 4 α α H ∣ α ∣ 2 + 4 α ( α H α ) α H ∣ α ∣ 4 = I n − 4 α α H ∣ α ∣ 2 + 4 α ( ∣ α ∣ 2 ) α H ∣ α ∣ 4 = I n − 4 α α H ∣ α ∣ 2 + 4 α α H ∣ α ∣ 2 = I n 2. A H A = ( I n − 2 α α H ∣ α ∣ 2 ) H ( I n − 2 α α H ∣ α ∣ 2 ) = ( I n − 2 α α H ∣ α ∣ 2 ) ( I n − 2 α α H ∣ α ∣ 2 ) = A 2 = I ∴ A 是 U 阵 \begin{aligned} 1.A^2&=(I_n-\frac{2\alpha\alpha^H}{\vert \alpha \vert^2})(I_n-\frac{2\alpha\alpha^H}{\vert \alpha \vert^2})\\ &=I_n^2-\frac{4\alpha\alpha^H}{\vert \alpha \vert^2}+\frac{4(\alpha\alpha^H)(\alpha\alpha^H)}{\vert \alpha\vert^4}=I_n-\frac{4\alpha\alpha^H}{\vert \alpha \vert^2}+\frac{4\alpha(\alpha^H\alpha)\alpha^H}{\vert \alpha\vert^4} \\ &=I_n-\frac{4\alpha\alpha^H}{\vert \alpha \vert^2}+\frac{4\alpha(\vert \alpha\vert^2)\alpha^H}{\vert \alpha\vert^4}=I_n-\frac{4\alpha\alpha^H}{\vert \alpha \vert^2}+\frac{4\alpha\alpha^H}{\vert \alpha \vert^2}=I_n\\ 2.A^HA&=(I_n-\frac{2\alpha\alpha^H}{\vert \alpha \vert^2})^H(I_n-\frac{2\alpha\alpha^H}{\vert \alpha \vert^2})\\ &=(I_n-\frac{2\alpha\alpha^H}{\vert \alpha \vert^2})(I_n-\frac{2\alpha\alpha^H}{\vert \alpha \vert^2})=A^2=I\\ &\therefore A是U阵 \end{aligned} 1.A22.AHA=(In−∣α∣22ααH)(In−∣α∣22ααH)=In2−∣α∣24ααH+∣α∣44(ααH)(ααH)=In−∣α∣24ααH+∣α∣44α(αHα)αH=In−∣α∣24ααH+∣α∣44α(∣α∣2)αH=In−∣α∣24ααH+∣α∣24ααH=In=(In−∣α∣22ααH)H(In−∣α∣22ααH)=(In−∣α∣22ααH)(In−∣α∣22ααH)=A2=I∴A是U阵
eg :
α = ( 1 1 1 ) , 其 U 阵为 I 3 − 2 α α H ∣ α ∣ 2 = ( 1 0 0 0 1 0 0 0 1 ) − 2 3 ( 1 1 1 1 1 1 1 1 1 ) = ( 1 3 − 2 3 − 2 3 − 2 3 1 3 − 2 3 − 2 3 − 2 3 1 3 ) \begin{aligned} \alpha=\left( \begin{matrix} 1\\1\\1 \end{matrix} \right),其U阵为 I_3-\frac{2\alpha\alpha^H}{\vert \alpha\vert^2}&=\left( \begin{matrix} 1&0&0\\0&1&0\\0&0&1\\ \end{matrix} \right)-\frac{2}{3}\left( \begin{matrix} 1&1&1\\1&1&1\\1&1&1\\ \end{matrix} \right)\\ &=\left( \begin{matrix} \frac{1}{3}&-\frac{2}{3}&-\frac{2}{3}\\ -\frac{2}{3}&\frac{1}{3}&-\frac{2}{3}\\ -\frac{2}{3}&-\frac{2}{3}&\frac{1}{3}\\ \end{matrix} \right) \end{aligned} α= 111 ,其U阵为I3−∣α∣22ααH= 100010001 −32 111111111 = 31−32−32−3231−32−32−3231
定义: A H = A A^H=A AH=A ,则矩阵为 A A A
证明:
A = ( a 11 ∗ a 22 ⋱ ∗ a n n ) A=\left( \begin{matrix} a_{11}&\quad&\quad &*\\ \quad&a_{22}&\quad&\quad \\ \quad &\quad&\ddots&\quad\\ *&\quad&\quad&a_{nn} \end{matrix} \right) A= a11∗a22⋱∗ann ,而 A H = ( a 11 ‾ ∗ a 22 ‾ ⋱ ∗ a n n ‾ ) A^H=\left( \begin{matrix} \overline{a_{11}}&\quad&\quad &*\\ \quad&\overline{a_{22}}&\quad&\quad \\ \quad &\quad&\ddots&\quad\\ *&\quad&\quad&\overline{a_{nn}} \end{matrix} \right) AH= a11∗a22⋱∗ann ,由Hermite性质, A H = A A^H=A AH=A ,则 a 11 = a 11 ‾ , a 22 = a 22 ‾ , . . . , a n n = a n n ‾ a_{11}=\overline{a_{11}},a_{22}=\overline{a_{22}},...,a_{nn}=\overline{a_{nn}} a11=a11,a22=a22,...,ann=ann ,可见 Hermite阵对角线元素为实数
若 A H = A A^H=A AH=A 是Hermite矩阵,则特征根都是实数, { λ 1 , ⋯ , λ n } ∈ R \{\lambda_1,\cdots,\lambda_n\}\in R {λ1,⋯,λn}∈R
由特商公式 λ = X H A X ∣ X ∣ 2 ,其中 X ≠ 0 ,为 λ 的特征向量 其中 ∣ X ∣ ≥ 0 ,为实数 . ∴ 若证 λ 为实数,即证 X H A X 为实数 已知 X H A X 为一维数字,则只需证明 ( X H A X ) H = X H A X 即可 , 已知 A = A H 为 H e r m i t e 矩阵, ( X H A X ) H = X H A H X = X H A X ∈ R \begin{aligned} &由特商公式\lambda=\frac{X^HAX}{\vert X \vert^2},其中X\neq 0,为\lambda的特征向量\\ &其中\vert X \vert\ge 0 ,为实数.\\ &\therefore 若证\lambda 为实数,即证 X^HAX为实数\\ &已知X^HAX为一维数字,则只需证明(X^HAX)^H=X^HAX即可,\\ &已知A=A^H为Hermite矩阵,(X^HAX)^H=X^HA^HX=X^HAX\in R \end{aligned} 由特商公式λ=∣X∣2XHAX,其中X=0,为λ的特征向量其中∣X∣≥0,为实数.∴若证λ为实数,即证XHAX为实数已知XHAX为一维数字,则只需证明(XHAX)H=XHAX即可,已知A=AH为Hermite矩阵,(XHAX)H=XHAHX=XHAX∈R
若 A = A H ∈ C n × n A=A^H\in C^{n\times n} A=AH∈Cn×n ,则 A A A 有 n n n 个互相正交的特征向量,即 X 1 ⊥ X 2 ⊥ . . . ⊥ X n X_1\bot X_2\bot...\bot X_n X1⊥X2⊥...⊥Xn
若 A 为 H e r m i t e 阵,则 ∃ U 阵 Q ,使 Q − 1 A Q = Q H A Q = Λ = ( λ 1 ⋱ λ n ) 令 Q = ( X 1 , X 2 , ⋯ , X n ) ,且 X 1 ⊥ X 2 ⊥ . . . ⊥ X n , A Q = Q Λ ⟺ ( A X 1 A X 2 ⋱ A X n ) = ( λ 1 X 1 λ 2 X 2 ⋱ λ n X n ) 即 X i 为矩阵 A 的 λ i 的特征向量 \begin{aligned} &若A为Hermite阵,则\exist U阵Q,使Q^{-1}AQ=Q^HAQ=\Lambda=\left( \begin{matrix} \lambda_1&&\\ &\ddots&\\ &&\lambda_n \end{matrix} \right)\\ &令Q=(X_1,X_2,\cdots,X_n),且X_1\bot X_2\bot...\bot X_n,\\ &AQ=Q\Lambda\iff \left( \begin{matrix} AX_1&&\\ &AX_2&&\\ &&\ddots&\\ &&&AX_n \end{matrix} \right)=\left( \begin{matrix} \lambda_1X_1&&&\\ &\lambda_2X_2&&\\ &&\ddots&\\ &&&\lambda_nX_n \end{matrix} \right)\\ &即X_i为矩阵A的\lambda_i的特征向量 \end{aligned} 若A为Hermite阵,则∃U阵Q,使Q−1AQ=QHAQ=Λ= λ1⋱λn 令Q=(X1,X2,⋯,Xn),且X1⊥X2⊥...⊥Xn,AQ=QΛ⟺ AX1AX2⋱AXn = λ1X1λ2X2⋱λnXn 即Xi为矩阵A的λi的特征向量
推论
若 A A A 为Hermite阵 A H = A A^H=A AH=A ,且 λ 1 ≠ λ 2 \lambda_1 \neq \lambda_2 λ1=λ2 ,则相应的特征向量正交
证明:
由 λ 1 ≠ λ 2 , 且 A X 1 = λ 1 X 1 , A X 2 = λ 2 X 2 , 有 λ 1 ( X 1 , X 2 ) = ( λ 1 X 1 , X 2 ) = ( A X 1 , X 2 ) = X 2 H A X 1 λ 2 ( X 1 , X 2 ) = ( X 1 , λ 2 ‾ X 2 ) = H 阵特征值 ∈ R ( X 1 , λ 2 X 2 ) = ( X 1 , A X 2 ) = ( A X 2 ) H X 1 = = X 2 H A X 1 = λ 1 ( X 1 , X 2 ) ∴ ( λ 1 − λ 2 ) ( X 1 , X 2 ) = 0 , 而 λ 1 ≠ λ 2 ,则 ( X 1 , X 2 ) = 0 \begin{aligned} &由\lambda_1 \neq \lambda_2,且AX_1=\lambda_1 X_1,AX_2=\lambda_2X_2,有\\ &\lambda_1(X_1,X_2)=(\lambda_1X_1,X_2)=(AX_1,X_2)=X_2^HAX_1\\ &\lambda_2(X_1,X_2)=(X_1,\overline{\lambda_2}X_2)\xlongequal{H阵特征值\in R}(X_1,\lambda_2X_2)=(X_1,AX_2)=(AX_2)^HX_1=\\ &\quad\quad =X_2^HAX_1=\lambda_1(X_1,X_2)\\ &\therefore (\lambda_1-\lambda_2)(X_1,X_2)=0,而\lambda_1\neq \lambda_2,则(X_1,X_2)=0 \end{aligned} 由λ1=λ2,且AX1=λ1X1,AX2=λ2X2,有λ1(X1,X2)=(λ1X1,X2)=(AX1,X2)=X2HAX1λ2(X1,X2)=(X1,λ2X2)H阵特征值∈R(X1,λ2X2)=(X1,AX2)=(AX2)HX1==X2HAX1=λ1(X1,X2)∴(λ1−λ2)(X1,X2)=0,而λ1=λ2,则(X1,X2)=0
若 A = A H A=A^H A=AH 是Hermite阵,则存在优阵 Q Q Q ,使 Q − 1 A Q = Q H A Q = Λ = ( λ 1 ⋱ λ n ) Q^{-1}AQ=Q^HAQ=\Lambda=\left( \begin{matrix} \lambda_1&\quad&\quad\\ \quad&\ddots&\quad\\ \quad&\quad&\lambda_n \end{matrix} \right) Q−1AQ=QHAQ=Λ= λ1⋱λn A = Q Λ Q − 1 = Q Λ Q H A=Q\Lambda Q^{-1} = Q\Lambda Q^H A=QΛQ−1=QΛQH ,且 λ ( A ) ∈ R \lambda(A)\in R λ(A)∈R
由许尔公式 ⇒ 有 U 阵 Q 使 Q − 1 A Q = Q H A Q = B = ( λ 1 ∗ ⋯ ∗ 0 λ 2 ⋯ ∗ ⋮ ⋮ ⋱ ⋮ 0 0 ⋯ λ n ) 由 A 是 H e r m i t e 矩阵,则 A H = A , ( Q H A Q ) H = Q H A H Q = Q H A Q = ( λ 1 ‾ 0 ⋯ 0 ∗ λ 2 ‾ ⋯ 0 ⋮ ⋮ ⋱ ⋮ ∗ ∗ ⋯ λ n ‾ ) 由此可见, B 为对角阵,且 λ i 为实数 \begin{aligned} &由许尔公式\Rightarrow 有U阵Q使Q^{-1}AQ=Q^HAQ=B\\ &=\left( \begin{matrix} \lambda_1&*&\cdots&*\\ 0&\lambda_2&\cdots&*\\ \vdots&\vdots &\ddots&\vdots\\ 0&0 &\cdots&\lambda_n\\ \end{matrix} \right)\\ &由A是Hermite矩阵,则A^H=A,(Q^HAQ)^H=Q^HA^HQ=Q^HAQ\\ &=\left( \begin{matrix} \overline{\lambda_1}&0&\cdots&0\\ *&\overline{\lambda_2}&\cdots&0\\ \vdots&\vdots &\ddots&\vdots\\ *&* &\cdots&\overline{\lambda_n}\\ \end{matrix} \right)\\ &由此可见,B为对角阵,且 \lambda_i 为实数 \end{aligned} 由许尔公式⇒有U阵Q使Q−1AQ=QHAQ=B= λ10⋮0∗λ2⋮0⋯⋯⋱⋯∗∗⋮λn 由A是Hermite矩阵,则AH=A,(QHAQ)H=QHAHQ=QHAQ= λ1∗⋮∗0λ2⋮∗⋯⋯⋱⋯00⋮λn 由此可见,B为对角阵,且λi为实数
任一矩阵 A n × p A_{n\times p} An×p , A H A A^HA AHA 与 A A H AA^H AAH 都是 H e r m i t e Hermite Hermite 矩阵
( A H A ) H = A H A , ( A A H ) H = A \begin{aligned} &(A^HA)^H=A^HA,(AA^H)^H=A \end{aligned} (AHA)H=AHA,(AAH)H=A
X = ( x 1 x 2 ⋮ x n ) , X H = ( x 1 ‾ , x 2 ‾ , ⋯ , x n ‾ ) t r ( X X H ) = t r ( X H X ) = ∣ x 1 ∣ 2 + ∣ x 2 ∣ 2 + ⋯ ∣ x n ∣ 2 = ∑ ∣ x j ∣ 2 \begin{aligned} &X=\left( \begin{matrix} x_1\\ x_2\\\vdots \\x_n \end{matrix} \right),X^H=(\overline{x_1},\overline{x_2},\cdots,\overline{x_n})\\ &tr(XX^H)=tr(X^HX)=\mid x_1 \mid^2+\mid x_2 \mid^2+\cdots\mid x_n \mid^2=\sum \mid x_j \mid^2 \end{aligned} X= x1x2⋮xn ,XH=(x1,x2,⋯,xn)tr(XXH)=tr(XHX)=∣x1∣2+∣x2∣2+⋯∣xn∣2=∑∣xj∣2
t r ( X Y H ) = t r ( Y H X ) = x 1 y 1 ‾ + ⋯ + x n y n ‾ = ∑ i = 1 n x i y i ‾ tr(XY^H)=tr(Y^HX)=x_1\overline{y_1}+\cdots+x_n\overline{y_n} =\sum\limits_{i=1}\limits^{n}x_i\overline{y_i} tr(XYH)=tr(YHX)=x1y1+⋯+xnyn=i=1∑nxiyi , X X X 为列向量
A n × p = ( a 11 ⋯ a 1 p ⋮ ⋱ ⋮ a n 1 ⋯ a n p ) ∈ C n × p A_{n\times p}=\left( \begin{matrix} a_{11}&\cdots&a_{1p}\\ \vdots&\ddots&\vdots\\ a_{n1}&\cdots&a_{np} \end{matrix} \right)\in C^{n\times p} An×p= a11⋮an1⋯⋱⋯a1p⋮anp ∈Cn×p , A p × n H = ( a 11 ‾ ⋯ a n 1 ‾ ⋮ ⋱ ⋮ a 1 p ‾ ⋯ a n p ‾ ) ∈ C p × n A^H_{p\times n}=\left( \begin{matrix} \overline{a_{11}}&\cdots&\overline{a_{n1}}\\ \vdots&\ddots&\vdots\\ \overline{a_{1p}}&\cdots&\overline{a_{np}} \end{matrix} \right)\in C^{p\times n} Ap×nH= a11⋮a1p⋯⋱⋯an1⋮anp ∈Cp×n
A A H = ( a 11 ⋯ a 1 p ⋮ ⋱ ⋮ a n 1 ⋯ a n p ) ( a 11 ‾ ⋯ a n 1 ‾ ⋮ ⋱ ⋮ a 1 p ‾ ⋯ a n p ‾ ) = ( a 11 a 11 ‾ + a 12 a 12 ‾ + ⋯ + a 1 p a 1 p ‾ ∗ a 21 a 21 ‾ + a 22 a 22 ‾ + ⋯ + a 2 p a 2 p ‾ ⋱ ∗ a n 1 a n 1 ‾ + a n 2 a n 2 ‾ + ⋯ + a n p a n p ‾ ) \begin{aligned} &AA^H=\left( \begin{matrix} a_{11}&\cdots&a_{1p}\\ \vdots&\ddots&\vdots\\ a_{n1}&\cdots&a_{np} \end{matrix} \right)\left( \begin{matrix} \overline{a_{11}}&\cdots&\overline{a_{n1}}\\ \vdots&\ddots&\vdots\\ \overline{a_{1p}}&\cdots&\overline{a_{np}} \end{matrix} \right)\\ &=\left( \begin{matrix} a_{11}\overline{a_{11}}+a_{12}\overline{a_{12}}+\cdots+a_{1p}\overline{a_{1p}}&\quad &\quad&\ast\\ \quad&a_{21}\overline{a_{21}}+a_{22}\overline{a_{22}}+\cdots+a_{2p}\overline{a_{2p}}&\quad&\quad\\ \quad &\quad &\ddots&\quad\\ \ast &\quad&\quad&a_{n1}\overline{a_{n1}}+a_{n2}\overline{a_{n2}}+\cdots+a_{np}\overline{a_{np}} \end{matrix} \right) \end{aligned} AAH= a11⋮an1⋯⋱⋯a1p⋮anp a11⋮a1p⋯⋱⋯an1⋮anp = a11a11+a12a12+⋯+a1pa1p∗a21a21+a22a22+⋯+a2pa2p⋱∗an1an1+an2an2+⋯+anpanp
t r ( A H A ) = t r ( A A H ) = ( ∣ a 11 ∣ 2 + ∣ a 12 ∣ 2 + . . . + ∣ a 1 p ∣ 2 ) + ( ∣ a 21 ∣ 2 + ∣ a 22 ∣ 2 + . . . + ∣ a 2 p ∣ 2 ) + . . . + ( ∣ a n 1 ∣ 2 + ∣ a n 2 ∣ 2 + . . . + ∣ a n p ∣ 2 ) = ∑ i = 1 , j = 1 n ∣ a i j ∣ 2 \begin{aligned} &tr(A^HA)=tr(AA^H)=\\ &(\mid a_{11} \mid^2+\mid a_{12} \mid^2+...+\mid a_{1p} \mid^2)+(\mid a_{21} \mid^2+\mid a_{22} \mid^2+...+\mid a_{2p} \mid^2)+\\ &...+(\mid a_{n1} \mid^2+\mid a_{n2} \mid^2+...+\mid a_{np} \mid^2) =\sum\limits_{i=1,j=1}\limits^{n}\mid a_{ij} \mid^2 \end{aligned} tr(AHA)=tr(AAH)=(∣a11∣2+∣a12∣2+...+∣a1p∣2)+(∣a21∣2+∣a22∣2+...+∣a2p∣2)+...+(∣an1∣2+∣an2∣2+...+∣anp∣2)=i=1,j=1∑n∣aij∣2
推论
t r ( A B H ) = t r ( B H A ) = ∑ a i j b i j ‾ tr(AB^H)=tr(B^HA)=\sum a_{ij}\overline{b_{ij}} tr(ABH)=tr(BHA)=∑aijbij
将 A、B 矩阵按列分块,可验证 t r ( B H A ) tr(B^HA) tr(BHA)
A = ( α 1 , α 2 ) , B = ( β 1 , β 2 ) , B H = ( β 1 ‾ T β 2 ‾ T ) B H A = ( β 1 ‾ T β 2 ‾ T ) ( α 1 , α 2 ) = ( β 1 ‾ T α 1 β 1 ‾ T α 2 β 2 ‾ T α 1 β 2 ‾ T α 2 ) t r ( B H A ) = β 1 ‾ T α 1 + β 2 ‾ T α 2 = ( a 11 b 11 ‾ + a 21 b 21 ‾ + a 31 b 31 ‾ ) + ( a 12 b 12 ‾ + a 22 b 22 ‾ + a 32 b 32 ‾ ) = ∑ a i j b i j ‾ \begin{aligned} &A=(\alpha_1,\alpha_2),B=(\beta_1,\beta_2),B^H=\left( \begin{matrix} \overline{\beta_1}^T\\ \overline{\beta_2}^T\\ \end{matrix} \right)\\ &B^HA=\left( \begin{matrix} \overline{\beta_1}^T\\ \overline{\beta_2}^T\\ \end{matrix} \right)(\alpha_1,\alpha_2)=\left( \begin{matrix} &\overline{\beta_1}^T\alpha_1\quad &\overline{\beta_1}^T\alpha_2\\ &\overline{\beta_2}^T\alpha_1\quad &\overline{\beta_2}^T\alpha_2 \end{matrix} \right)\\ &tr(B^HA)=\overline{\beta_1}^T\alpha_1+\overline{\beta_2}^T\alpha_2=\\ &\quad (a_{11}\overline{b_{11}}+a_{21}\overline{b_{21}}+a_{31}\overline{b_{31}})+ (a_{12}\overline{b_{12}}+a_{22}\overline{b_{22}}+a_{32}\overline{b_{32}})\\ &=\sum a_{ij}\overline{b_{ij}} \end{aligned} A=(α1,α2),B=(β1,β2),BH=(β1Tβ2T)BHA=(β1Tβ2T)(α1,α2)=(β1Tα1β2Tα1β1Tα2β2Tα2)tr(BHA)=β1Tα1+β2Tα2=(a11b11+a21b21+a31b31)+(a12b12+a22b22+a32b32)=∑aijbij
将 A、B 矩阵按行分块,可验证 t r ( A B H ) tr(AB^H) tr(ABH)
A = ( α 1 α 2 α 3 ) , B = ( β 1 β 2 β 3 ) , B H = ( β 1 ‾ T , β 2 ‾ T , β 3 ‾ T ) A B H = ( α 1 α 2 α 3 ) ( β 1 ‾ T , β 2 ‾ T , β 3 ‾ T ) = ( α 1 β 1 ‾ T α 1 β 2 ‾ T α 1 β 3 ‾ T α 2 β 1 ‾ T α 2 β 2 ‾ T α 3 β 3 ‾ T α 3 β 1 ‾ T α 3 β 2 ‾ T α 3 β 3 ‾ T ) t r ( A B H ) = ( a 11 b 11 ‾ + a 12 b 12 ‾ ) + ( a 21 b 21 ‾ + a 22 b 22 ‾ ) + ( a 31 b 31 ‾ + a 32 b 32 ‾ ) = ∑ a i j b i j ‾ \begin{aligned} &A=\left( \begin{matrix} \alpha_1\\ \alpha_2\\ \alpha_3 \end{matrix} \right), B=\left( \begin{matrix} \beta_1\\ \beta_2\\ \beta_3 \end{matrix} \right),B^H=(\overline{\beta_1}^T,\overline{\beta_2}^T,\overline{\beta_3}^T)\\ &AB^H=\left( \begin{matrix} \alpha_1\\ \alpha_2\\ \alpha_3 \end{matrix} \right)(\overline{\beta_1}^T,\overline{\beta_2}^T,\overline{\beta_3}^T)=\left( \begin{matrix} \alpha_1\overline{\beta_1}^T\quad&\alpha_1\overline{\beta_2}^T\quad&\alpha_1\overline{\beta_3}^T\\ \alpha_2\overline{\beta_1}^T\quad&\alpha_2\overline{\beta_2}^T\quad&\alpha_3\overline{\beta_3}^T\\ \alpha_3\overline{\beta_1}^T\quad&\alpha_3\overline{\beta_2}^T\quad&\alpha_3\overline{\beta_3}^T \end{matrix} \right)\\ &tr(AB^H)=(a_{11}\overline{b_{11}}+a_{12}\overline{b_{12}})+(a_{21}\overline{b_{21}}+a_{22}\overline{b_{22}})+(a_{31}\overline{b_{31}}+a_{32}\overline{b_{32}})\\ &\quad\quad\quad\quad = \sum a_{ij}\overline{b_{ij}} \end{aligned} A= α1α2α3 ,B= β1β2β3 ,BH=(β1T,β2T,β3T)ABH= α1α2α3 (β1T,β2T,β3T)= α1β1T