Skip to content

6.3-exam3-2010-b

Quiz 3, 2010 · Section 6.3: Symmetric Positive Definite Matrices Source: exam3_2.pdf, pp. 8-9, 12-13

Problem

3. (35 points)

(a) (15 pts.) The matrix \(A\) has independent columns. The matrix \(C\) is square, diagonal, and has positive entries. Why is the matrix \(K = A^TCA\) positive definite? You can use any of the basic tests for positive definiteness.

(b) (20 pts.) If a diagonalizable matrix \(A\) has orthonormal eigenvectors and real eigenvalues must it be symmetric? (Briefly why or give a counterexample)

Solution (a) Again, there are many ways to do this. First way: remember that a symmetric matrix is positive definite if and only if it can be written as \(R^TR\) where \(R\) has linearly independent columns. In our case, let \(c_1, \dots, c_n\) be the diagonal entries of \(C\) and let \(B\) be the diagonal matrix with diagonal entries \(\sqrt{c_1}, \dots, \sqrt{c_n}\) (take the positive square roots). Then \(C = B^TB\) and so \(K = A^TB^TBA = (BA)^T(BA)\), so we take \(R = BA\). Since \(A\) has linearly independent columns and \(B\) is invertible (because the \(c_i\) are nonzero numbers), we conclude that \(BA\) also has linearly independent columns. Second way: use the energy definition. Let \(x\) be a nonzero vector. We have to show that \(x^TKx > 0\). First, since \(A\) has linearly independent columns, this means that its nullspace is 0, so \(Ax \neq 0\). Set \(y = Ax\). Since \(C\) is diagonal and has positive diagonal entries, it is positive definite (this follows from the eigenvalue definition, or the submatrices definition, for example). So \(y^TCy > 0\), but \(y^TCy = x^TKx\), so we're done. (b) True, since \(A\) is diagonalizable with real eigenvalues, we can write \(A = S\Lambda S^{-1}\) where \(\Lambda\) has real entries and the columns of \(S\) are some eigenvectors. Since we also know that \(A\) has orthonormal eigenvectors, we may choose these for \(S\), and hence \(S^{-1} = S^T\). But then \(A = S\Lambda S^T\) and \(A^T = (S^T)^T\Lambda^TS^T\). But \((S^T)^T = S\) for any matrix \(S\), and \(\Lambda^T = \Lambda\) because it is diagonal and square. So \(A = A^T\) and \(A\) is symmetric. Remark: some people said that Hermitian matrices (topic not covered) are diagonalizable with orthonormal eigenvectors and have real eigenvalues but are not symmetric in general. This is true, and technically did not violate the directions of the problem since it did not specify that \(A\) has to have real entries, so received full credit.

Keywords: positive definite, energy test, symmetric matrix, orthonormal eigenvectors