跳到主要内容

東京大学 新領域創成科学研究科 メディカル情報生命専攻 2024年1月実施 問題8

Author

zephyr, 祭音Myyura

Description

Suppose that the eigenvalues and the corresponding eigenvectors of an n×nn \times n square matrix A\mathbf{A} are λ1,,λn\lambda_1, \dots, \lambda_n and α1,,αn\mathbf{\alpha}_1, \dots, \mathbf{\alpha}_n respectively.

Suppose that In\mathbf{I}_n is the n×nn \times n identity matrix, and the inverse matrix of an invertible matrix C\mathbf{C} is C1\mathbf{C}^{-1}.

Answer the following questions.

  1. Show all the eigenvalues and the corresponding eigenvectors of A2\mathbf{A}^2.

  2. If λ1,,λn\lambda_1, \dots, \lambda_n are mutually different, show that P1AP\mathbf{P}^{-1} \mathbf{A} \mathbf{P} is a diagonal matrix, using P=(α1,,αn)\mathbf{P} = (\mathbf{\alpha}_1, \dots, \mathbf{\alpha}_n) that is a matrix of concatenated eigenvectors.

  3. Show all the eigenvalues and the corresponding eigenvectors of B\mathbf{B}.

B=(300232001)\mathbf{B} = \begin{pmatrix} 3 & 0 & 0 \\ -2 & 3 & 2 \\ 0 & 0 & 1 \end{pmatrix}
  1. Suppose that μ\mu is the maximum eigenvalue of B\mathbf{B}, and γ=(100)\mathbf{\gamma} = \begin{pmatrix} 1 \\ 0 \\ 0 \end{pmatrix}. Calculate δ=(BμI3)γ\mathbf{\delta} = (\mathbf{B} - \mu \mathbf{I}_3)\mathbf{\gamma}.

  2. Suppose that β\beta is the eigenvector of B\mathbf{B} corresponding to the minimum eigenvalue. Calculate Q1BQ\mathbf{Q}^{-1}\mathbf{B}\mathbf{Q} using Q=(δ,γ,β)\mathbf{Q} = (\mathbf{\delta}, \mathbf{\gamma}, \beta) that is a matrix concatenating δ,γ,β\mathbf{\delta}, \mathbf{\gamma}, \beta.

  3. Suppose that mm is an arbitrary positive integer. Calculate Bm\mathbf{B}^m.


假设 n×nn \times n 方阵 A\mathbf{A} 的特征值及相应的特征向量分别为 λ1,,λn\lambda_1, \dots, \lambda_nα1,,αn\mathbf{\alpha}_1, \dots, \mathbf{\alpha}_n

假设 In\mathbf{I}_nn×nn \times n 的单位矩阵,并且可逆矩阵 C\mathbf{C} 的逆矩阵为 C1\mathbf{C}^{-1}

回答以下问题。

  1. 展示 A2\mathbf{A}^2 的所有特征值及相应的特征向量。

  2. 如果 λ1,,λn\lambda_1, \dots, \lambda_n 是互不相同的,证明 P1AP\mathbf{P}^{-1} \mathbf{A} \mathbf{P} 是一个对角矩阵,其中 P=(α1,,αn)\mathbf{P} = (\mathbf{\alpha}_1, \dots, \mathbf{\alpha}_n) 是由特征向量构成的矩阵。

  3. 展示 B\mathbf{B} 的所有特征值及相应的特征向量。

B=(300232001)\mathbf{B} = \begin{pmatrix} 3 & 0 & 0 \\ -2 & 3 & 2 \\ 0 & 0 & 1 \end{pmatrix}
  1. 假设 μ\muB\mathbf{B} 的最大特征值,并且 γ=(100)\mathbf{\gamma} = \begin{pmatrix} 1 \\ 0 \\ 0 \end{pmatrix}
    计算 δ=(BμI3)γ\mathbf{\delta} = (\mathbf{B} - \mu \mathbf{I}_3)\mathbf{\gamma}

  2. 假设 β\betaB\mathbf{B} 对应于最小特征值的特征向量。计算 Q1BQ\mathbf{Q}^{-1}\mathbf{B}\mathbf{Q},其中 Q=(δ,γ,β)\mathbf{Q} = (\mathbf{\delta}, \mathbf{\gamma}, \beta) 是由 δ,γ,β\mathbf{\delta}, \mathbf{\gamma}, \beta 构成的矩阵。

  3. 假设 mm 是任意正整数。计算 Bm\mathbf{B}^m

题目描述

n×nn\times n 方阵 A\mathbf A 的特征值及对应特征向量分别为

λ1,,λn,α1,,αn.\lambda_1,\ldots,\lambda_n,\qquad \boldsymbol\alpha_1,\ldots,\boldsymbol\alpha_n.

In\mathbf I_n 表示单位矩阵,可逆矩阵 C\mathbf C 的逆记为 C1\mathbf C^{-1}。回答:

  1. 列出 A2\mathbf A^2 的全部特征值及对应特征向量。

  2. λ1,,λn\lambda_1,\ldots,\lambda_n 两两不同,令

    P=(α1,,αn),\mathbf P=(\boldsymbol\alpha_1,\ldots,\boldsymbol\alpha_n),

    证明 P1AP\mathbf P^{-1}\mathbf A\mathbf P 为对角矩阵。

  3. B=(300232001)\mathbf B= \begin{pmatrix} 3&0&0\\ -2&3&2\\ 0&0&1 \end{pmatrix}

    的全部特征值与相应特征向量。

  4. μ\muB\mathbf B 的最大特征值,

    γ=(100),\boldsymbol\gamma=\begin{pmatrix}1\\0\\0\end{pmatrix},

    计算

    δ=(BμI3)γ.\boldsymbol\delta=(\mathbf B-\mu\mathbf I_3)\boldsymbol\gamma.
  5. β\boldsymbol\betaB\mathbf B 最小特征值对应的特征向量,并令

    Q=(δ,γ,β),\mathbf Q=(\boldsymbol\delta,\boldsymbol\gamma,\boldsymbol\beta),

    计算 Q1BQ\mathbf Q^{-1}\mathbf B\mathbf Q

  6. 对任意正整数 mm,计算 Bm\mathbf B^m

Kai

1. Eigenpairs of A2\mathbf A^2

For every ii,

A2αi=A(λiαi)=λi2αi.\mathbf A^2\boldsymbol\alpha_i =\mathbf A(\lambda_i\boldsymbol\alpha_i) =\lambda_i^2\boldsymbol\alpha_i.

Thus the eigenvalues of A2\mathbf A^2 are λi2\lambda_i^2 (with algebraic multiplicity), and each αi\boldsymbol\alpha_i is a corresponding eigenvector. If A\mathbf A is diagonalizable, the full eigenspace is

Eμ(A2)=λσ(A)λ2=μEλ(A).E_\mu(\mathbf A^2)= \bigoplus_{\substack{\lambda\in\sigma(\mathbf A)\\\lambda^2=\mu}} E_\lambda(\mathbf A).

Without diagonalizability, the listed eigenvectors need not describe every eigenvector of A2\mathbf A^2.

2. Diagonalization of A\mathbf A

Distinct eigenvalues have linearly independent eigenvectors, so P\mathbf P is invertible. Since

AP=(λ1α1,,λnαn)=Pdiag(λ1,,λn),\mathbf A\mathbf P =(\lambda_1\boldsymbol\alpha_1,\ldots,\lambda_n\boldsymbol\alpha_n) =\mathbf P\operatorname{diag}(\lambda_1,\ldots,\lambda_n),

we obtain

P1AP=diag(λ1,,λn).\mathbf P^{-1}\mathbf A\mathbf P =\operatorname{diag}(\lambda_1,\ldots,\lambda_n).

3. Eigenpairs of B\mathbf B

det(BλI3)=(3λ)2(1λ).\det(\mathbf B-\lambda\mathbf I_3)=(3-\lambda)^2(1-\lambda).

Hence

λ=3:E3=span ⁣{(010)},λ=1:E1=span ⁣{(011)}.\lambda=3:\quad E_3=\operatorname{span}\!\left\{\begin{pmatrix}0\\1\\0\end{pmatrix}\right\}, \qquad \lambda=1:\quad E_1=\operatorname{span}\!\left\{\begin{pmatrix}0\\-1\\1\end{pmatrix}\right\}.

The eigenvalue 33 has algebraic multiplicity 22 and geometric multiplicity 11.

4. Computing δ\boldsymbol\delta

Here μ=3\mu=3, so

δ=(B3I3)γ=(020).\boldsymbol\delta=(\mathbf B-3\mathbf I_3)\boldsymbol\gamma =\begin{pmatrix}0\\-2\\0\end{pmatrix}.

5. Computing Q1BQ\mathbf Q^{-1}\mathbf B\mathbf Q

Choose β=(0,1,1)T\boldsymbol\beta=(0,-1,1)^T. Then

Bδ=3δ,Bγ=δ+3γ,Bβ=β.\mathbf B\boldsymbol\delta=3\boldsymbol\delta, \qquad \mathbf B\boldsymbol\gamma=\boldsymbol\delta+3\boldsymbol\gamma, \qquad \mathbf B\boldsymbol\beta=\boldsymbol\beta.

Therefore

Q1BQ=(310030001).\mathbf Q^{-1}\mathbf B\mathbf Q =\begin{pmatrix} 3&1&0\\ 0&3&0\\ 0&0&1 \end{pmatrix}.

6. Computing Bm\mathbf B^m

Raising the Jordan block above to the mm-th power gives

Bm=(3m002m3m13m3m1001)(m1).\boxed{ \mathbf B^m= \begin{pmatrix} 3^m&0&0\\ -2m3^{m-1}&3^m&3^m-1\\ 0&0&1 \end{pmatrix}} \qquad(m\ge1).

Knowledge

重点词汇

  • singular value decomposition (SVD) 奇异值分解
  • pseudoinverse 广义逆
  • surjective 满射
  • injective 单射
  • orthogonal decomposition 正交分解

参考资料

  1. "Linear Algebra and Its Applications" by Gilbert Strang, Chapter 7: The Singular Value Decomposition (SVD)
  2. "Matrix Computations" by Gene H. Golub and Charles F. Van Loan, Chapter 2: Matrix Analysis