Skip to content

6.3-exam3-2005

Quiz 3, 2005 · Section 6.3: Symmetric Positive Definite Matrices Source: exam3_1.pdf, pp. 83, 90-91

Problem

3 (30 pts.) Suppose \(A\) is a 3 by 3 symmetric matrix with eigenvalues 2, 5, 7 and corresponding eigenvectors \(x_1, x_2, x_3\).

(a) Suppose \(x\) is a combination \(x = c_1x_1 + c_2x_2 + c_3x_3\). Find \(Ax\). Now find \(x^TAx\) using the symmetry of \(A\). Prove that \(x^TAx > 0\) (unless \(x = 0\)).

(b) Suppose those eigenvectors have length 1 (unit vectors). Show that \(B = 2x_1x_1^T + 5x_2x_2^T + 7x_3x_3^T\) has the same eigenvectors and eigenvalues as \(A\). Is \(B\) necessarily the same matrix as \(A\) (yes or no)?

(c) For which numbers \(b\) does this matrix have 3 positive eigenvalues?

\[ A = \begin{bmatrix} 2 & b & 3 \\ b & 2 & b \\ 3 & b & 4 \end{bmatrix} \]

Note: The SVD will be on the final when you have more time to digest it.

Solution (a) We write \[ Ax = c_1Ax_1 + c_2Ax_2 + c_3Ax_3 = 2c_1x_1 + 5c_2x_2 + 7c_3x_3 \] and \[ x^TAx = (c_1x_1^T + c_2x_2^T + c_3x_3^T)(2c_1x_1 + 5c_2x_2 + 7c_3x_3) = 2c_1^2x_1^Tx_1 + 5c_2^2x_2^Tx_2 + 7c_3^2x_3^Tx_3 \] (opening the parentheses, we use the fact that eigenvectors of a symmetric matrix corresponding to different eigenvalues are orthogonal, and hence \(x_i^Tx_j = 0\) for \(i \neq j\)). Since \(x_i^Tx_i = \|x_i\|^2 > 0\) and \(c_i^2 > 0\) unless \(c_i = 0\), we conclude that \(x^TAx > 0\) unless \(c_1 = c_2 = c_3 = 0\), i.e. \(x = 0\). (b) We have \[ Bx_1 = 2x_1x_1^Tx_1 + 5x_2x_2^Tx_1 + 7x_3x_3^Tx_1 = 2x_1 \] because \(x_1^Tx_1 = \|x_1\|^2 = 1\) and \(x_i^Tx_j = 0\) for \(i \neq j\). Thus \(x_1\) is an eigenvector of \(B\) with eigenvalue \(\lambda_1 = 2\). Similarly, we can show that \(Bx_2 = 5x_2\) and \(Bx_3 = 7x_3\). Since both \(A\) and \(B\) have the diagonalization \[ \begin{bmatrix} x_1 & x_2 & x_3 \end{bmatrix} \begin{bmatrix} 2 & & \\ & 5 & \\ & & 7 \end{bmatrix} \begin{bmatrix} x_1 & x_2 & x_3 \end{bmatrix}^{-1}, \] they are the same matrix. (c) \(A\) has 3 positive eigenvalues if and only if it is positive definite. To test for positive definiteness, we check the three upper-left determinants to see when they are positive. The 1 by 1 upper-left determinant is 2, which is positive. The 2 by 2 upper-left determinant is \(4 - b^2\), which is positive whenever \(-2 < b < 2\). Finally, we compute the 3 by 3 upper-left determinant, or \(\det A\): \[ \det A = 2 \det \begin{bmatrix} 2 & b \\ b & 4 \end{bmatrix} - b \det \begin{bmatrix} b & b \\ 3 & 4 \end{bmatrix} + 3 \det \begin{bmatrix} b & 2 \\ 3 & b \end{bmatrix} = 2(8 - b^2) - b(4b - 3b) + 3(b^2 - 6) = -2, \] which is always negative. Since \(\det A < 0\) regardless of the value of \(b\), we conclude that \(A\) cannot have 3 positive eigenvalues.

Keywords: symmetric matrix, eigenvalues, eigenvectors, positive definite, determinant