Skip to content

6.3-exam3-2008-strang

Quiz 3, 2008 (strang) · Section 6.3: Symmetric Positive Definite Matrices Source: exam3_1.pdf, pp. 144-145

Problem

1 (40 pts.) The (real) matrix \(A\) is

\[ A = \begin{bmatrix} 1 & 1 & 2 \\ 1 & x & 3 \\ 2 & 3 & 6 \end{bmatrix}. \]

(a) What can you tell me about the eigenvectors of \(A\)? What is the sum of its eigenvalues?

(b) For which values of \(x\) is this matrix \(A\) positive definite?

(c) For which values of \(x\) is \(A^2\) positive definite? Why?

(d) If \(R\) is any rectangular matrix, prove from \(x^T(R^TR)x\) that \(R^TR\) is positive semidefinite (or definite). What condition on \(R\) is the test for \(R^TR\) to be positive definite?

Solution (a) Since \(A\) is a symmetric matrix (no matter what \(x\) is), its eigenvectors may be chosen orthonormal. The sum of the eigenvalues is the same as the trace of \(A\), that is, the sum of the diagonal entries: \(tr(A) = 7 + x\). (b) In this case, the easiest tests for positive definiteness are the pivot test and the determinant test. A matrix \(A\) is positive definite when every one of the top-left determinants is positive. In this case, the three determinants are 1, \(x - 1\), and \[ \det(A) = 1(6x - 9) - (6 - 6) + 2(3 - 2x) = 2x - 3. \] All of these are positive precisely when \(x > 3/2\). (c) Perhaps the clearest way to think about this is by using the eigenvalues. Suppose \(A\) has eigenvalues \(\lambda_1, \lambda_2, \lambda_3\). (They are all real because \(A\) is symmetric.) Then the eigenvalues of \(A^2\) are \(\lambda_1^2, \lambda_2^2, \lambda_3^2\). These are all positive so long as the eigenvalues are non-zero. So, \(A^2\) is positive definite except when \(A\) has an eigenvalue of 0, or equivalently, except when \(A\) is not invertible. We found in part (b) that \(\det(A) = 0\) only when \(x = 3/2\). Thus, the final answer is that \(A^2\) is positive definite except when \(x = 3/2\). One could also find \(A^2\) explicitly and use the determinant or pivot test. However, you could notice that the top left entry of \(A^2\) is 6, the 2 by 2 determinant is \(6(10 + x^2) - (7 + x)^2 = 5x^2 - 14x + 11 > 0\), and the 3 by 3 determinant is \(\det(A^2) = \det(A)^2\). The only way that any of these could be non-positive is if \(\det(A) = 0\). (d) We use the \(x^T Ax\) test for positive (semi)definiteness. We have \[ x^T R^T Rx = (Rx)^T Rx = Rx \cdot Rx. \] This is just the length of the vector \(Rx\). This length is positive when \(Rx\) is not the zero vector and is 0 when \(Rx\) is the 0 vector. In particular, since this number is always at least 0, \(R^T R\) is definitely positive semidefinite. It is positive definite when this number is positive for any nonzero \(x\). That is, we need for \(Rx\) to only be the 0 vector when \(x\) is the 0 vector. This is equivalent to saying that \(R\) has trivial nullspace, or \(R\) has full column rank.

Keywords: positive definite, positive semidefinite, eigenvalues, trace, determinant