线性代数 05.04 对称矩阵的相似矩阵

$\color{blue}{\S 第五章 第四节 对称矩阵的相似矩阵}$

$定理5.对称矩阵的特征值为实数.$
$证:设复数\lambda为对称矩阵A的特征值,\\ 复向量x为\lambda对应的特征向量,\\ 即Ax = \lambda x, x \neq 0. 用\bar \lambda 表示\lambda 的共轭复数,\\ 用\bar x表示x的共轭复向量. \\ 则 A \bar x = \bar A \bar x = (\overline {Ax}) = (\overline {\lambda x}) = \bar \lambda \bar x.于是有\\ \bar x^T Ax = \bar x^T (Ax) = \bar x^T \lambda x = \lambda \bar x^T x, \\ 又因为A是对称矩阵, A = A^T \\ \bar x^T A x = (\bar x^T A^T)x = (A \bar x)^Tx = (\bar \lambda \bar x) x = \bar \lambda \bar x^T x. \\ 两式相减,得 \\ (\lambda - \bar \lambda) \bar x^T x = 0, \\ 但因x \neq 0, 所以\\ \bar x^T x = \sum_{i=1}^{n} \bar x_i x_i = \sum_{i = 1} ^{n} |x_i|^2 \neq 0, \\\ 故 \lambda - \bar \lambda = 0,即 \lambda = \bar \lambda,这就说明\lambda 是实数.$

$显然,当特征值\lambda_i为实数时,齐次线性方程组 \\ (A - \lambda E)x = 0 \\ 是实系数方程组,由 |A - \lambda_i E | = 0知必有实的基础解系,\\ 所以对应的特征向量可以取实向量.$

$定理6.设\lambda_1, \lambda_2是对称矩阵A的两个特征值,\\ p_1,p_2是对应的特征值,若\lambda_1 \neq \lambda_2,则p_1与p_2正交.$
$证:因为Ap_1 = \lambda_1 p_1,Ap_2 = \lambda_2 p_2,且\lambda_1 \neq \lambda_2, \\ 所以,\lambda_1p_2^Tp_1 = p_2^T \lambda_1 p_1 = p_2^T A p_1 = p_2^TA^Tp_1 \\ = (Ap_2)^Tp_1 = (\lambda_2 p_2)^T p_1 = \lambda_2p_2^Tp_1 \\ 从而，(\lambda_1 - \lambda_2)p_2^Tp_1 = 0 \\ 但 \lambda_1 \neq \lambda_2,故 p_2^Tp_1 = 0 \\ 即p_1与p_2正交.$

$定理7.设A为n阶对称矩阵,\lambda是A的特征方程的r重根,\\ 则矩阵A - \lambda E的秩R(A - \lambda E) = n - r,从而对应\\ 特征值\lambda恰有r个线性无关的特征向量.$

$定理8.设A为n阶的对称矩阵,则必有正交矩阵P,\\ 使P^{-1}AP = \Lambda,其中\Lambda是以A的n个特征值为\\ 对角元素的对角矩阵.$
$证:设A的互不相等的特征值为\lambda_1, \lambda_2, \cdots, \lambda_s, \\ 它们的重数依次为r_1, r_2, \cdots, r_s(r_1 + r_2 + \cdots + r_s = n). \\ 根据定理5及定理7知,对应特征值\lambda_i(i = 1, 2, \cdots, s),\\ 恰有r_i个线性无关的实特征向量,把它们施密特标准正交化,\\ 即得r_i个两两正交单位特征向量.由r_1 + r_2 + \cdots + r_s = n,\\ 知这样的特征向量共可得n个. \\ 按定理6知对应于不同特征值的特征向量正交,故这n个单位特征\\ 向量两两正交.于是以它们为列向量构成正交矩阵P,并有\\ P^{-1}AP = \Lambda. \\ 其中对角矩阵\Lambda的对角元素含r_1个\lambda_1, r_2个\lambda_2, \cdots, r_s个\lambda_s, \\ 恰是A的n个特征值.$

$例1.设 A = \begin{pmatrix} 4 & 0 & 0 \\ 0 & 3 & 1 \\ 0 & 1 & 3 \\ \end{pmatrix}, \\ 求一个正交矩阵P,使P^{-1}AP = \Lambda为对角矩阵.$
$解:①由|A - \lambda E| = 0,求A的全部特征值. |A - \lambda E| = \begin{vmatrix} 4 - \lambda & 0 & 0 \\ 0 & 3 - \lambda & 1 \\ 0 & 1 & 3 - \lambda \\ \end{vmatrix} \\ = (-1)^{(1 + 1)}(4 - \lambda)[(3 - \lambda)(3 - \lambda) - 1 \times 1] \\ = -(\lambda - 2)(\lambda -4)^2 = 0 \\ 解得A的特征值为:\lambda_1 = 2, \lambda_2 = \lambda_3 = 4. \\ ②由 (A - \lambda E)x = 0,求A的特征向量. \\ 当\lambda_1 = 2时, (A - 2E)x = 0,由\\ A - 2E = \begin{pmatrix} 2 & 0 & 0 \\ 0 & 1 & 1 \\ 0 & 1 & 1 \\ \end{pmatrix} \sim \begin{pmatrix} 1 & 0 & 0 \\ 0 & 1 & 1 \\ 0 & 0 & 0 \\ \end{pmatrix} \\ 得基础解系 p_1 = \begin{pmatrix} 0 \\ -1 \\ 1 \end{pmatrix},单位化p_1 = \dfrac{1}{\sqrt{2}} \begin{pmatrix} 0 \\ -1 \\ 1 \end{pmatrix} \\ 当\lambda_2 = \lambda_3 = 4时, 由\\ A - 4E = \begin{pmatrix} 0 & 0 & 0 \\ 0 & -1 & 1 \\ 0 & 1 & -1 \\ \end{pmatrix} \sim \begin{pmatrix} 0 & 0 & 0 \\ 0 & 1 & -1 \\ 0 & 0 & 0 \\ \end{pmatrix} \\ 解得基础解系 p_2 = \begin{pmatrix} 1 \\ 0 \\ 0 \end{pmatrix}, p_3 = \begin{pmatrix} 0 \\ 1 \\ 1 \end{pmatrix} \\ 基础解系中两个向量恰好正交,单位化得两个正交的单位特征向量.\\ 单位化p_2 = \begin{pmatrix} 1 \\ 0 \\ 0 \end{pmatrix}, p_3 = \dfrac{1}{\sqrt{2}} \begin{pmatrix} 0 \\ 1 \\ 1 \end{pmatrix} \\ ③ 由p_1,p_2,p_3拼出正交矩阵P, \\ P = (p_1, p_2, p_3) = \begin{pmatrix} 0 & 1 & 0 \\ -\dfrac{1}{\sqrt{2}} & 0 & \dfrac{1}{\sqrt{2}} \\ \dfrac{1}{\sqrt{2}} & 0 & \dfrac{1}{\sqrt{2}} \\ \end{pmatrix} \\ ④可以验证确有 \\ P^{-1}AP = P^TAP = \begin{pmatrix} 2 & 0 & 0 \\ 0 & 4 & 0 \\ 0 & 0 & 4 \\ \end{pmatrix}$
$在此例中对应于\lambda = 4,若求得方程(A - \lambda E) x = 0的基础解系 \\ \alpha_1 = \begin{pmatrix} 1 \\ 1 \\ 1 \end{pmatrix}, \alpha_2 = \begin{pmatrix} -1 \\ 1 \\ 1 \end{pmatrix}, \\ 则需要把它施密特标准正交化:\\ 令b_1 = \alpha_1, e_1 = \dfrac{b_1}{\Vert b_1 \Vert} = \dfrac{1}{\sqrt{3}} \begin{pmatrix} 1 \\ 1 \\ 1 \end{pmatrix}, \\ 令b_2 = \alpha_2 - [\alpha_2, e_1]e_1 \\ = \alpha_2 - [\alpha_2, b_1]\dfrac{b_1}{\Vert b_1 \Vert ^2} \\ = \begin{pmatrix} -1 \\ 1 \\ 1 \end{pmatrix} - \dfrac{1}{3}\begin{pmatrix} 1 \\ 1 \\ 1 \end{pmatrix} \\ = \begin{pmatrix} -\dfrac{4}{3} \\ \dfrac{2}{3} \\ \dfrac{2}{3} \end{pmatrix} \\ = \dfrac{2}{3}\begin{pmatrix} -2 \\ 1 \\ 1 \end{pmatrix}, \\ e_2 = \dfrac{b_2}{\Vert b_2 \Vert } = \dfrac{1}{\sqrt{6}}\begin{pmatrix} -2 \\ 1 \\ 1 \end{pmatrix}. 于是得正交矩阵 \\ p = \begin{pmatrix} 0 & \dfrac{1}{\sqrt{3}} & -\dfrac{2}{\sqrt{6}} \\ -\dfrac{1}{\sqrt{2}} & \dfrac{1}{\sqrt{3}} & \dfrac{1}{\sqrt{6}} \\ \dfrac{1}{\sqrt{2}} & \dfrac{1}{\sqrt{3}} & \dfrac{1}{\sqrt{6}} \\ \end{pmatrix} \\ 可以验证确有 \\ P^{-1}AP = P^TAP = \begin{pmatrix} 2 & 0 & 0 \\ 0 & 4 & 0 \\ 0 & 0 & 4 \\ \end{pmatrix} \\ 由此可见,{\color{green}{把一个方阵化成对角矩阵所用的正交变换矩阵P不是唯一的.}}$

$例2.设矩阵A = \begin{pmatrix} 1 & 0 & 1 \\ 0 & 2 & 0 \\ 1 & 0 & 1 \\ \end{pmatrix}, \\ 矩阵B = (kE + A)^2,其中k为实数,E为单位矩阵,\\ 求对角矩阵\Lambda,使得B与\Lambda相似,并求k为何值时,B为奇异矩阵.$
$解:由 \\ |A - \lambda E| = \begin{vmatrix} 1 - \lambda & 0 & 1 \\ 0 & 2 - \lambda & 0 \\ 1 & 0 & 1 - \lambda \\ \end{vmatrix} \\ = (-1)^{(2 + 2)}(2 - \lambda)[(1 - \lambda)^2 - 1 \times 1] \\ = -\lambda(\lambda -2)^2 = 0, \\ 解得A的特征值为:\lambda_1 = 0, \lambda_2 = \lambda_3 = 2. \\ 记对角矩阵 \Lambda = \begin{pmatrix} 0 & 0 & 0 \\ 0 & 2 & 0 \\ 0 & 0 & 2 \\ \end{pmatrix}, \\ 因为A为对称矩阵,故存在正交矩阵P,使得P^TAP = \Lambda. \\ 所以 A = P \Lambda P^T,于是 \\ B = (kE + A)^2 = (kPP^T + P\Lambda P^T)^2 \\ = [ P(kE + \Lambda)P^T) ][ P(kE + \Lambda)P^T) ] \\ = P (kE + \Lambda)^2P^T \\ = P \begin{pmatrix}k^2 & & \\ & (k+2)^2 & \\ & & (k+2)^2 \end{pmatrix} P^T \\ 由此可得B对应相似对角矩阵的\\ \Lambda = \begin{pmatrix} k^2 & & \\ & (k+2)^2 & \\ & & (k+2)^2 \\ \end{pmatrix}. \\ 显然,B与A相似,且k = 0或k = -2时, B为奇异矩阵.$

$例3.设3阶方阵A的特征值为\\ \lambda_1 = 1, \lambda_2 = 2, \lambda_3 = 3,对应的特征向量依次为\\ \alpha_1 = (1, 1, 1)^T, \alpha_2 = (1, 2, 4)^T, \alpha_3 = (1, 3, 9)^T, \\ 又向量\beta = (1, 1, 3)^T. \\ (1) 将\beta用\alpha_1, \alpha_2, \alpha_3线性表示. \\ (2) 求A^n\beta.$
$解:\\ (1) 设 \beta = x_1 \alpha_1 + x_2 \alpha_2 + x_3 \alpha_3, 即\\ \begin{pmatrix} 1 & 1 & 1 \\ 1 & 2 & 3 \\ 1 & 4 & 9 \\ \end{pmatrix} \begin{pmatrix} x_1 \\ x_2 \\ x_3 \\ \end{pmatrix} = \begin{pmatrix} 1 \\ 1 \\ 3 \\ \end{pmatrix}, \\ 由于增广矩阵 \begin{pmatrix} 1 & 1 & 1 & 1 \\ 1 & 2 & 3 & 1 \\ 1 & 4 & 9 & 3 \\ \end{pmatrix} \sim \begin{pmatrix} 1 & 0 & 0 & 2 \\ 0 & 1 & 0 & -2 \\ 0 & 0 & 1 & 1 \\ \end{pmatrix} \\ 解得 x_1 = 2, x_2 = -2, x_3 = 1,故 \\ \beta = 2\alpha_1 - 2\alpha_2 + \alpha_3. \\ (2) 令 \Lambda = \begin{pmatrix} 1 & & \\ & 2 & \\ & & 3 \\ \end{pmatrix}, P = (\alpha_1, \alpha_2, \alpha_3) = \begin{pmatrix} 1 & 1 & 1 \\ 1 & 2 & 3 \\ 1 & 4 & 9 \\ \end{pmatrix}, \\ 则A = P \Lambda P^{-1},即A^n = P \Lambda ^n P^{-1}, \\ 而\beta = (\alpha_1, \alpha_2, \alpha_3)\begin{pmatrix} 2 \\ -2 \\ 1 \\ \end{pmatrix} = P \alpha, \\ 从而 A^n\beta = P \Lambda^n P^{-1} \beta \\ = P \Lambda^n P^{-1} P \alpha \\ = P \Lambda^n \alpha \\ = \begin{pmatrix} 1 & 1 & 1 \\ 1 & 2 & 3 \\ 1 & 4 & 9 \\ \end{pmatrix}\begin{pmatrix} 1 & & \\ & 2^n & \\ & & 3^n \\ \end{pmatrix}\begin{pmatrix} 2 \\ -2 \\ 1 \\ \end{pmatrix} \\ = \begin{pmatrix} 1 & 2^n & 3^n \\ 1 & 2^{n+1} & 3^{n+1} \\ 1 & 2^{n+2} & 3^{n+2} \\ \end{pmatrix}\begin{pmatrix} 2 \\ -2 \\ 1 \\ \end{pmatrix} \\ = \begin{pmatrix} 2 & -2^{n+1} & 3^n \\ 2 & -2^{n+2} & 3^{n+1} \\ 2 & -2^{n+3} & 3^{n+2} \\ \end{pmatrix}$