Skip to content

7.1-final-2015-b

Final Exam, 2015 · Section 7.1: Singular Values and Singular Vectors Source: finalexam_2.pdf, p. 156

Problem

4 (5+5 pts.) Symmetric positive definite matrices \(H\) and orthogonal matrices \(Q\) are the most important. Here is a great theorem: Every square invertible matrix A can be factored into A = HQ.

(a) Start from \(A = U\Sigma V^T\) (the SVD) and choose \(Q = UV^T\). Find the other factor \(H\) so that \(U\Sigma V^T = HQ\). Why is your \(H\) symmetric and why is it positive definite?

(b) Factor this 2 by 2 matrix into \(A = U\Sigma V^T\) and then into \(A = HQ\):

\[ A = \begin{bmatrix} 1 & 3 \\ -1 & 3 \end{bmatrix} = U\Sigma V^T = HQ. \]
Solution (a) From \(A = HQ = HUV^T\) we solve for \(H\), inverting the orthogonal matrices: \[ H = U\Sigma V^T (UV^T)^{-1} = U\Sigma V^T V U^T = U\Sigma U^T. \] This is symmetric: \[ H^T = (U\Sigma U^T)^T = U\Sigma^T U^T = U\Sigma U^T = H, \] since \(\Sigma\) is diagonal. It is positive definite by the eigenvalue test: the eigenvalues of \(H\) are the diagonal entries of \(\Sigma\), i.e. the singular values of \(A\), which are the nonnegative eigenvalues of \(A^TA\); since \(A\) is invertible they are all strictly positive. (b) Compute \[ A^T A = \begin{bmatrix} 1 & -1 \\ 3 & 3 \end{bmatrix} \begin{bmatrix} 1 & 3 \\ -1 & 3 \end{bmatrix} = \begin{bmatrix} 2 & 0 \\ 0 & 18 \end{bmatrix}. \] So the singular values are \(\sigma_1 = \sqrt{18} = 3\sqrt{2}\) and \(\sigma_2 = \sqrt{2}\), with eigenvectors \(v_1 = (0, 1)\), \(v_2 = (1, 0)\): \[ V = \begin{bmatrix} 0 & 1 \\ 1 & 0 \end{bmatrix}. \] Then \[ u_1 = \frac{Av_1}{\sigma_1} = \begin{bmatrix} 1/\sqrt{2} \\ 1/\sqrt{2} \end{bmatrix}, \qquad u_2 = \frac{Av_2}{\sigma_2} = \begin{bmatrix} 1/\sqrt{2} \\ -1/\sqrt{2} \end{bmatrix}, \] so the SVD is \[ A = \begin{bmatrix} 1/\sqrt{2} & 1/\sqrt{2} \\ 1/\sqrt{2} & -1/\sqrt{2} \end{bmatrix} \begin{bmatrix} 3\sqrt{2} & 0 \\ 0 & \sqrt{2} \end{bmatrix} \begin{bmatrix} 0 & 1 \\ 1 & 0 \end{bmatrix}. \] Finally, \[ H = U\Sigma U^T = \begin{bmatrix} 2\sqrt{2} & \sqrt{2} \\ \sqrt{2} & 2\sqrt{2} \end{bmatrix}, \qquad Q = UV^T = \begin{bmatrix} 1/\sqrt{2} & 1/\sqrt{2} \\ -1/\sqrt{2} & 1/\sqrt{2} \end{bmatrix}, \] and one checks that \(HQ = A\).

Keywords: SVD, polar decomposition, symmetric positive definite, orthogonal matrix