Skip to content

7.1-exam3-2011

Quiz 3, 2011 · Section 7.1: Singular Values and Singular Vectors Source: exam3_2.pdf, pp. 21-22

Problem

2. (30 pts.)

  1. (5 pts) An orthogonal matrix \(Q\) satisfies \(Q^TQ = QQ^T = I\). What are the \(n\) singular values of \(Q\)?

  2. (10 pts) Let \(A = \begin{bmatrix} 1 & & \\ & -2 & \\ & & 3 \end{bmatrix}\). Find an SVD, meaning \(A = U\Sigma V^T\), where \(U\) and \(V\) are orthogonal, and \(\Sigma = \begin{bmatrix} \sigma_1 & & \\ & \sigma_2 & \\ & & \sigma_3 \end{bmatrix}\) is diagonal with \(\sigma_1 \geq \sigma_2 \geq \sigma_3 \geq 0\). (Be sure that the factorization is correct and satisfies all stated requirements.)

  3. (15 pts) The \(2 \times 2\) matrix \(A = \sigma_1 u_1 v_1^T + \sigma_2 u_2 v_2^T\), where \(\sigma_1 > \sigma_2 > 0\) and both \(u_1, u_2\) and \(v_1, v_2\) are orthonormal bases for \(\mathbb{R}^2\).

The set of all vectors \(x\) with \(\|x\| = 1\) describes a circle in the plane. What shape best describes the set of all vectors \(Ax\) with \(\|x\| = 1\)? Draw a general picture of that set labeling all the relevant quantities \(\sigma_1, \sigma_2, u_1, u_2\) and \(v_1, v_2\). (Hint: Why are \(u_1, u_2\) relevant and \(v_1, v_2\) not relevant?)

Solution 1. They are all 1. The singular values are the positive square roots of the eigenvalues of \(QQ^T = Q^TQ = I\). 2. \[ A = \begin{bmatrix} & & 1 \\ & -1 & \\ 1 & & \end{bmatrix}\begin{bmatrix} 3 & & \\ & 2 & \\ & & 1 \end{bmatrix}\begin{bmatrix} & & 1 \\ & 1 & \\ 1 & & \end{bmatrix}. \] The singular values are in decreasing order and are positive. One can compute \(AA^T\) and \(A^TA\), but it is easier to rig the permutation matrices and correct the sign. 3. An ellipse with axes labeled \(\sigma_1u_1\) and \(\sigma_2u_2\). The SVD rotates (or reflects) the circle with \(V^T\), scales it to an ellipse with axes in the coordinate directions through \(\Sigma\), and then yields a rotated ellipse with axes in the directions \(u_1\) and \(u_2\) after \(U\) is applied. The matrix \(\Sigma\) scales the \(x\) and \(y\) axes by \(\sigma_1\) and \(\sigma_2\) respectively, and \(\sigma_1\) is the longer of the two.

Keywords: singular value decomposition, singular values, orthogonal matrix, ellipse