7.1-exam3-2006¶
Quiz 3, 2006 · Section 7.1: Singular Values and Singular Vectors Source: exam3_1.pdf, pp. 100, 107-108
Problem¶
1 (34 pts.) (a) If a square matrix \(A\) has all \(n\) of its singular values equal to 1 in the SVD, what basic classes of matrices does \(A\) belong to? (Singular, symmetric, orthogonal, positive definite or semidefinite, diagonal)
(b) Suppose the (orthonormal) columns of \(H\) are eigenvectors of \(B\):
The eigenvalues of \(B\) are \(\lambda = 0, 1, 2, 3\). Write \(B\) as the product of 3 specific matrices. Write \(C = (B + I)^{-1}\) as the product of 3 matrices.
(c) Using the list in question (a), which basic classes of matrices do \(B\) and \(C\) belong to? (Separate question for \(B\) and \(C\))
Solution
(a) If \(\Sigma = I\) then \(A = UV^{T}\), a product of orthogonal matrices, so \(A\) is an orthogonal matrix. Second proof: all \(\sigma_i = 1\) implies \(A^{T}A = I\). So \(A\) is orthogonal. (\(A\) is *never* singular, and it won't always be symmetric — take \(U = \begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}\) and \(V = I\), for example. This also shows it can't be diagonal, or positive definite or semidefinite.) (b) \(B = H\Lambda H^{-1}\) with \[ \Lambda = \begin{bmatrix} 0 & & & \\ & 1 & & \\ & & 2 & \\ & & & 3 \end{bmatrix}. \] Similarly, \[ (B + I)^{-1} = H(\Lambda + I)^{-1}H^{-1} \quad \text{with} \quad (\Lambda + I)^{-1} = \begin{bmatrix} 1 & & & \\ & 1/2 & & \\ & & 1/3 & \\ & & & 1/4 \end{bmatrix}, \] the same eigenvectors \(H\). (c) \(B\) is singular, symmetric, positive semidefinite. \(C\) is symmetric positive definite.Keywords: singular values, orthogonal matrix, SVD, diagonalization, positive semidefinite