Skip to content

7.1-final-2008-strang

Final Exam, 2008 (strang) · Section 7.1: Singular Values and Singular Vectors Source: finalexam_1.pdf, p. 219

Problem

8 (8 pts.) Every invertible matrix \(A\) is the product \(A = QH\) of an orthogonal matrix \(Q\) and a symmetric positive definite matrix \(H\). I will start the proof:

\(A\) has a singular value decomposition \(A = U\Sigma V^T\).

Then \(A = (UV^T)(V\Sigma V^T)\).

(a) Show that \(UV^T\) is an orthogonal matrix \(Q\) (what is the test for an orthogonal matrix?).

(b) Show that \(V\Sigma V^T\) is a symmetric positive definite matrix. What are its eigenvalues and eigenvectors? Why did I need to assume that \(A\) is invertible?

Solution (a) To test that \(Q = UV^T\) is orthogonal, we must show that \(Q^TQ = I\). But \[ Q^TQ = (UV^T)^TUV^T = VU^TUV^T = V(I)V^T = I. \] We used the fact that \(U\) and \(V\) are orthogonal matrices. (b) The matrix \(H = V\Sigma V^T\) is definitely symmetric, as \(H^T = V\Sigma^TV^T = V\Sigma V^T\) because \(\Sigma\) is diagonal. Furthermore, note that the expression \(H = V\Sigma V^T\) is a diagonalization of \(H\). This means that \(H\) has eigenvalues given by the entries of \(\Sigma\) and eigenvectors equal to the columns of \(V\). To show that \(H\) is positive-definite, we just need to show that the diagonal entries of \(\Sigma\) are all positive. Now, we know that they are all non-negative, because the SVD always gives us non-negative singular values. We must also show that none of the singular values are zero. Remember that the singular values are equal to the square roots of the eigenvalues of \(A^TA\). However, because \(A\) is invertible, the matrix \(A^TA\) is also invertible, and so can't have any eigenvalues equal to 0. So no singular value is 0 either.

Keywords: singular value decomposition, orthogonal matrix, positive definite, polar decomposition