Skip to content

7.1-final-2014-strang

Final Exam, 2014 · Section 7.1: Singular Values and Singular Vectors Source: finalexam_2.pdf, p. 141

Problem

7 (10 points) Every invertible matrix \(A\) equals an orthogonal matrix \(Q\) times a positive definite matrix \(S\). This famous fact comes directly from the SVD for the square matrix \(A = U\Sigma V^T\), by choosing \(Q = UV^T\).

(a) How can you prove that \(Q = UV^T\) is orthogonal?

(b) Substitute \(Q^{-1}\) and \(A\) to write \(S = Q^{-1}A\) in terms of \(U, V\) and \(\Sigma\). How can you tell that this matrix \(S\) is symmetric positive definite?

Solution (a) Since \(A\) is invertible, \(U\) and \(V\) are square orthogonal matrices of full rank. Then \[ Q^T Q = (UV^T)^T (UV^T) = V U^T U V^T = V I V^T = I, \] using \(U^T U = I\) and \(V V^T = I\) (orthogonality of \(U\) and \(V\)). Hence \(Q = UV^T\) is orthogonal. (b) Substituting, \[ S = Q^{-1} A = (UV^T)^{-1} U \Sigma V^T = V U^T U \Sigma V^T = V \Sigma V^T. \] This is symmetric, since \((V\Sigma V^T)^T = V \Sigma V^T\), and it is positive definite: writing \(\Sigma = \Sigma^{1/2}\Sigma^{1/2}\) with \(\Sigma^{1/2} = \mathrm{diag}(\sqrt{\sigma_i})\), \[ S = V \Sigma V^T = (\Sigma^{1/2}V^T)^T (\Sigma^{1/2}V^T), \] which is positive definite because \(\Sigma^{1/2}V^T\) is invertible (all \(\sigma_i > 0\)).

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