Skip to content

7.1-exam3-2011-strang

Quiz 3, 2011 (strang) · Section 7.1: Singular Values and Singular Vectors Source: exam3_2.pdf, p. 30

Problem

2. This problem is about the matrix

\[ A = \begin{bmatrix} \sqrt{2} & 1 \\ 0 & \sqrt{2} \end{bmatrix} \]

(a) Find all eigenvectors of \(A\). Exactly why is it impossible to diagonalize \(A\) in the form \(A = S\Lambda S^{-1}\)?

(b) Find the matrices \(U, \Sigma, V^T\) in the Singular Value Decomposition \(A = U\Sigma V^T\). Tell me two orthogonal vectors \(v_1, v_2\) in the plane so that \(Av_1\) and \(Av_2\) are also orthogonal.

(c) Find a matrix \(B\) that is similar to \(A\) (but different from \(A\)). Show that \(A\) and \(B\) meet the requirement to be similar (what is it?)

Solution (a) The characteristic polynomial is \[ p(\lambda) = (\lambda - \sqrt{2})(\lambda - \sqrt{2}) = 0, \] so \(\lambda_1 = \lambda_2 = \sqrt{2}\). Then \[ N(A - \sqrt{2}I) = N\left(\begin{bmatrix} 0 & 1 \\ 0 & 0 \end{bmatrix}\right) = \operatorname{span}\left\{ \begin{bmatrix} 1 \\ 0 \end{bmatrix} \right\}, \] so the only eigenvectors are multiples of \(\begin{bmatrix} 1 \\ 0 \end{bmatrix}\). There are not enough independent eigenvectors to form an invertible matrix \(S\) with eigenvectors as its columns, so it is impossible to diagonalize \(A\). (b) We compute the eigenvalues of \(A^TA\): \[ A^TA = \begin{bmatrix} \sqrt{2} & 0 \\ 1 & \sqrt{2} \end{bmatrix}\begin{bmatrix} \sqrt{2} & 1 \\ 0 & \sqrt{2} \end{bmatrix} = \begin{bmatrix} 2 & \sqrt{2} \\ \sqrt{2} & 3 \end{bmatrix}. \] \[ p_{A^TA}(\lambda) = \lambda^2 - 5\lambda + 4 = (\lambda - 1)(\lambda - 4) = 0 \quad \Longrightarrow \quad \lambda_1 = 4,\ \lambda_2 = 1. \] Thus \(\sigma_1 = \sqrt{\lambda_1} = 2\) and \(\sigma_2 = \sqrt{\lambda_2} = 1\), so \(\Sigma = \begin{bmatrix} 2 & 0 \\ 0 & 1 \end{bmatrix}\). The eigenvectors of \(A^TA\) are \[ N(A^TA - 4I) = N\begin{bmatrix} -2 & \sqrt{2} \\ \sqrt{2} & -1 \end{bmatrix} = \operatorname{span}\left\{ \begin{bmatrix} \sqrt{2} \\ 2 \end{bmatrix} \right\}, \qquad v_1 = \frac{1}{\sqrt{6}}\begin{bmatrix} \sqrt{2} \\ 2 \end{bmatrix}, \] \[ N(A^TA - I) = N\begin{bmatrix} 1 & \sqrt{2} \\ \sqrt{2} & 2 \end{bmatrix} = \operatorname{span}\left\{ \begin{bmatrix} \sqrt{2} \\ -1 \end{bmatrix} \right\}, \qquad v_2 = \frac{1}{\sqrt{3}}\begin{bmatrix} \sqrt{2} \\ -1 \end{bmatrix}. \] Then \[ u_1 = \frac{Av_1}{\sigma_1} = \frac{1}{\sqrt{3}}\begin{bmatrix} \sqrt{2} \\ 1 \end{bmatrix}, \qquad u_2 = \frac{Av_2}{\sigma_2} = \frac{1}{\sqrt{3}}\begin{bmatrix} 1 \\ -\sqrt{2} \end{bmatrix}. \] So \[ A = \begin{bmatrix} | & | \\ u_1 & u_2 \\ | & | \end{bmatrix}\begin{bmatrix} 2 & 0 \\ 0 & 1 \end{bmatrix}\begin{bmatrix} | & | \\ v_1 & v_2 \\ | & | \end{bmatrix}^T. \] The vectors \(Av_1\) and \(Av_2\) are orthogonal because \(u_1 \perp u_2\) (and \(Av_i = \sigma_iu_i\)). (c) We say \(B \sim A\) if \(B = MAM^{-1}\) for some invertible matrix \(M\). Choose, for example, \(M = \begin{bmatrix} 1 & 0 \\ 1 & 1 \end{bmatrix}\) (any invertible \(M\) works). Then \[ B = \begin{bmatrix} 1 & 0 \\ 1 & 1 \end{bmatrix}\begin{bmatrix} \sqrt{2} & 1 \\ 0 & \sqrt{2} \end{bmatrix}\begin{bmatrix} 1 & 0 \\ 1 & 1 \end{bmatrix}^{-1} = \begin{bmatrix} \sqrt{2} - 1 & 1 \\ -1 & 1 + \sqrt{2} \end{bmatrix}, \] and \(B \neq A\) but \(B \sim A\). (The printed solution gives \(B = \begin{bmatrix} \sqrt{2} & 1 \\ 1 & \sqrt{2} \end{bmatrix}\), but that matrix has eigenvalues \(\sqrt{2} \pm 1\) and so is not similar to \(A\); the computation above is the correct one.)

Keywords: singular value decomposition, singular values, eigenvectors, similar matrices, diagonalizable