G
N
I
D
A
O
L

正定矩阵的分解

正定矩阵的分解方法

设三阶正定矩阵 \(A\),若矩阵 \(A\) 的特征值为 \(\lambda_1,\lambda_2,\lambda_3\),对应的单位化特征向量分别为 \(\alpha_1,\alpha_2,\alpha_3\)两两正交,则存在正交矩阵 \(Q = (\alpha_1,\alpha_2,\alpha_3)\),使得 \(Q^{\mathrm{T}}AQ = \Lambda = \begin{bmatrix} \lambda_1 & & \\ & \lambda_2 & \\ & & \lambda_3 \end{bmatrix}\)

现得到 \(A = Q \Lambda Q^{\mathrm{T}}\),令 \(\Lambda_1 = \begin{bmatrix} \sqrt{\lambda_1} & & \\ & \sqrt{\lambda_2} & \\ & & \sqrt{\lambda_3} \end{bmatrix}\),则正定矩阵 \(A\) 有以下分解方法:

【方法一】 \(A = Q \Lambda Q^{\mathrm{T}} = Q (\Lambda_1 \Lambda_1) Q^{\mathrm{T}} = (\Lambda_1 Q^{\mathrm{T}})^{\mathrm{T}} (\Lambda_1 Q^{\mathrm{T}}) = C^{\mathrm{T}} C(令 C = \Lambda_1 Q^{\mathrm{T}})\)

【方法二】 \(A = Q \Lambda Q^{\mathrm{T}} = Q \Lambda_1 \Lambda_1 Q^{\mathrm{T}} = Q \Lambda_1 (Q^{\mathrm{T}} Q) \Lambda_1 Q^{\mathrm{T}} = (Q \Lambda_1 Q^{\mathrm{T}})(Q \Lambda_1 Q^{\mathrm{T}}) = C^2 (令 C = Q \Lambda_1 Q^{\mathrm{T}})\)

此时由谱分解定理得:

  • \(A = \lambda_1 \alpha_1 \alpha_1^{\mathrm{T}} + \lambda_2 \alpha_2 \alpha_2^{\mathrm{T}} + \lambda_3 \alpha_3 \alpha_3^{\mathrm{T}}\)
  • \(B = \sqrt{\lambda_1} \alpha_1 \alpha_1^{\mathrm{T}} + \sqrt{\lambda_2} \alpha_2 \alpha_2^{\mathrm{T}} + \sqrt{\lambda_3} \alpha_3 \alpha_3^{\mathrm{T}}\)

【拓展】 若要将正定矩阵 \(A\) 分解为 \(C^3\),则可令 \(\Lambda_2 = \begin{bmatrix} \sqrt[3]{\lambda_1} & & \\ & \sqrt[3]{\lambda_2} & \\ & & \sqrt[3]{\lambda_3} \end{bmatrix}\),所以有:\(A = (Q \Lambda_2 Q^{\mathrm{T}})(Q \Lambda_2 Q^{\mathrm{T}})(Q \Lambda_2 Q^{\mathrm{T}}) = C^3 (令 C = Q \Lambda_2 Q^{\mathrm{T}})\)


相关例题

【例 1】已知矩阵 \(A = \begin{bmatrix} 0 & 0 & 1 \\ 0 & 1 & 0 \\ 1 & 0 & 0 \end{bmatrix}\),求一个可逆矩阵 \(C\),使得 \(C^3 = A\)

【解】\(C^3 = A\) 的特征值为 \(1,1,-1\),则 \(C\) 的特征值为 \(\sqrt[3]{1},\sqrt[3]{1},\sqrt[3]{-1}\)(即 \(1,1,-1\)),所以 \(C-E\) 的特征值为 \(0,0,-2\)

\(A\) 的特征值 \(-1\) 所对应的特征向量为 \(\alpha_3 = \frac{1}{\sqrt{2}} (1,0,-1)^{\mathrm{T}}\),则 \(C-E\) 的特征值 \(\lambda_3 = -2\) 所对应的特征向量也为 \(\alpha_3 = \frac{1}{\sqrt{2}} (1,0,-1)^{\mathrm{T}}\)

由谱分解定理得

\[\begin{aligned} C-E &= \lambda_3 \alpha_3 \alpha_3^{\mathrm{T}} \\ &= -2 \cdot \frac{1}{\sqrt{2}} \begin{bmatrix} 1 \\ 0 \\ -1 \end{bmatrix} \cdot \frac{1}{\sqrt{2}} (1,0,-1) \\ &= -\begin{bmatrix} 1 & 0 & -1 \\ 0 & 0 & 0 \\ -1 & 0 & 1 \end{bmatrix} = \begin{bmatrix} -1 & 0 & 1 \\ 0 & 0 & 0 \\ 1 & 0 & -1 \end{bmatrix} \\ \therefore C &= (C-E) + E = \begin{bmatrix} 0 & 0 & 1 \\ 0 & 1 & 0 \\ 1 & 0 & 0 \end{bmatrix} \end{aligned} \]


【例 2】已知矩阵 \(A = \begin{bmatrix} 0 & 0 & 1 \\ 0 & 1 & 0 \\ 1 & 0 & 0 \end{bmatrix}\),求一个正定矩阵 \(C\),使得 \(C^2 = A+2E\)

【解】\(A\) 的特征值为 \(1,1,-1\),则 \(C^2 = A+2E\) 的特征值为 \(3,3,1\),则 \(C\) 的特征值为 \(\sqrt{3},\sqrt{3},1\),所以 \(C-\sqrt{3}E\) 的特征值为 \(0,0,1-\sqrt{3}\)

\(A\) 的特征值 \(-1\) 所对应的特征向量为 \(\alpha_3 = \frac{1}{\sqrt{2}} (1,0,-1)^{\mathrm{T}}\),则 \(C-\sqrt{3}E\) 的特征值 \(\lambda_3 = 1-\sqrt{3}\) 所对应的特征向量也为 \(\alpha_3 = \frac{1}{\sqrt{2}} (1,0,-1)^{\mathrm{T}}\)

由谱分解定理得

\[\begin{aligned} C-\sqrt{3}E &= \lambda_3 \alpha_3 \alpha_3^{\mathrm{T}} \\ &= (1-\sqrt{3}) \cdot \frac{1}{\sqrt{2}} \begin{bmatrix} 1 \\ 0 \\ -1 \end{bmatrix} \cdot \frac{1}{\sqrt{2}} (1,0,-1) \\ &= \frac{1-\sqrt{3}}{2} \begin{bmatrix} 1 & 0 & -1 \\ 0 & 0 & 0 \\ -1 & 0 & 1 \end{bmatrix} \\ &= \begin{bmatrix} \frac{1-\sqrt{3}}{2} & 0 & \frac{\sqrt{3}-1}{2} \\ 0 & 0 & 0 \\ \frac{\sqrt{3}-1}{2} & 0 & \frac{1-\sqrt{3}}{2} \end{bmatrix} \\ \therefore C &= (C-\sqrt{3}E) + \sqrt{3}E = \begin{bmatrix} \frac{\sqrt{3}+1}{2} & 0 & \frac{\sqrt{3}-1}{2} \\ 0 & \sqrt{3} & 0 \\ \frac{\sqrt{3}-1}{2} & 0 & \frac{\sqrt{3}+1}{2} \end{bmatrix} \end{aligned} \]


【例 3】已知矩阵 \(A = \begin{bmatrix} 0 & 0 & 1 \\ 0 & 1 & 0 \\ 1 & 0 & 0 \end{bmatrix}\),求一个正定矩阵 \(C\),使得 \(C^n = A+2E\)

【解】该题与上题的解法是相似的,现直接给出答案:

\[C = \begin{bmatrix} \frac{\sqrt[n]{3}+1}{2} & 0 & \frac{\sqrt[n]{3}-1}{2} \\ 0 & \sqrt[n]{3} & 0 \\ \frac{\sqrt[n]{3}-1}{2} & 0 & \frac{\sqrt[n]{3}+1}{2} \end{bmatrix} \]

posted @ 2023-08-29 19:08  漫舞八月(Mount256)  阅读(515)  评论(0编辑  收藏  举报