特征值分解与奇异值分解

1.特征值分解(谱分解)

A = Q Λ Q H { Q = [ u 1 ⃗ u 2 ⃗ ⋯ u M ⃗ ] Λ = d i a g { λ 1 , λ 2 , ⋯   , λ M } \begin{align} &A=Q\Lambda Q^H\\ &\left\{ \begin{aligned} Q&=\begin{bmatrix} \vec{u_1} &\vec{u_2} &\cdots&\vec{u_M} \end{bmatrix}\\ \Lambda&=diag\{\lambda_1,\lambda_2,\cdots,\lambda_M\} \end{aligned} \right. \end{align} ​A=QΛQH{QΛ​=[u1​ ​​u2​ ​​⋯​uM​ ​​]=diag{λ1​,λ2​,⋯,λM​}​​​

2.奇异值分解

A L × M = Y L × L [ Σ K × K 0 K × ( M − K ) 0 ( L − K ) × K 0 ( L − K ) × ( M − K ) ] X M × M H { X = [ x 1 ⃗ x 2 ⃗ ⋯ x M ⃗ ] Y = [ y 1 ⃗ y 2 ⃗ ⋯ y M ⃗ ] Σ = d i a g { σ 1 , σ 2 , ⋯   , σ M } \begin{align} &A_{L\times M}=Y_{L\times L} \begin{bmatrix} \Sigma_{K\times K } & 0_{K\times (M-K)}\\ 0_{(L-K)\times K}& 0_{(L-K)\times (M-K)} \end{bmatrix} X_{M\times M}^H \\ &\left\{ \begin{aligned} X&=\begin{bmatrix} \vec{x_1} &\vec{x_2} &\cdots&\vec{x_M} \end{bmatrix}\\ Y&=\begin{bmatrix} \vec{y_1} &\vec{y_2} &\cdots&\vec{y_M} \end{bmatrix}\\ \Sigma&=diag\{\sigma_1,\sigma_2,\cdots,\sigma_M\} \end{aligned} \right. \end{align} ​AL×M​=YL×L​[ΣK×K​0(L−K)×K​​0K×(M−K)​0(L−K)×(M−K)​​]XM×MH​⎩ ⎨ ⎧​XYΣ​=[x1​ ​​x2​ ​​⋯​xM​ ​​]=[y1​ ​​y2​ ​​⋯​yM​ ​​]=diag{σ1​,σ2​,⋯,σM​}​​​

3.特征值分解与奇异值分解关系

$$\begin{array}{r}\{\begin{array}{rl}{A}^{H}A& =X\left[\begin{array}{cc}{\Sigma }^2& 0_{K\times (M-K)}\\ 0_{(M-K)\times K}& 0_{(M-K)\times (M-K)}\end{array}\right]X^H\\ A{A}^H& =Y\left[\begin{array}{cc}{\Sigma }^2& 0_{K\times (L-K)}\\ 0_{(L-K)\times K}& 0_{(L-K)\times (L-K)}\end{array}\right]Y^H\end{array}\end{array}$$

(1)特征值与奇异值

$A^{H}A$与$A{A}^H$的非零特征值等于$A_{L\times M}$的非零奇异值的平方:
$$\Lambda ={\Sigma }^2$$

(2)特征向量与奇异向量

左奇异向量矩阵$Y$是$A{A}^H$的特征向量矩阵
左奇异向量矩阵$X$是$A^{H}A$的特征向量矩阵