6.2-exam3-1999-fall¶
Quiz 3, 1999 (fall) · Section 6.2: Diagonalizing a Matrix Source: exam3_1.pdf, pp. 36-37
Problem¶
2 (40 pts.) The matrix \(A\) has 3's on the diagonal and 2's everywhere else:
(a) Decide if \(A\) is positive definite. What is the minimum value of \(x^TAx\) for all vectors \(x\) in \(R^3\)?
(b) All entries of \(B = A - I\) are 2's. From its rank find all the eigenvalues of \(B\) and then all the eigenvalues of \(A\).
(c) Write down any one specific symmetric matrix \(C\) that is similar to \(A\). Write down if possible any one nonsymmetric matrix \(N\) that is similar to \(A\). Write down a matrix \(J\) with the same eigenvalues as \(A\) that is not similar to \(A\). (Give the 9 numbers in \(C, N, J\).)
(d) For the 6 by 6 matrix \(A_6\) with 3's on the diagonal and 2's everywhere else use the same method (with \(A_6 - I\)) to find the six eigenvalues. If you make a good choice of eigenvectors, in what form can you factor \(A\)?
Solution
(a) \(A\) is positive definite since \(A\) is symmetric and the upper left determinants are \(3, 5, 7\). \(A\) positive definite \(\Rightarrow x^TAx \geq 0\), and \(x^TAx = 0\) when \(x = 0\). So the minimum is 0. (b) All three columns of \(B\) are the same, so \(Rank(B) = 1\), which means \(N(B)\) has dimension 2, i.e. \(B\) has two independent eigenvectors with eigenvalue 0. Thus \(\lambda_1 = \lambda_2 = 0\), and \(\lambda_3 = tr(B) = 6\). The eigenvalues of \(A\) are therefore \(1, 1, 7\). (c) One valid choice is \[ C = \Lambda = \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 7 \end{bmatrix}, \qquad N = \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 1 \\ 0 & 0 & 7 \end{bmatrix}, \qquad J = \begin{bmatrix} 1 & 1 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 7 \end{bmatrix}. \] Explanation: \(A\) is symmetric (so you could have let \(C = A\)) and hence can be diagonalized to \(\Lambda\). Consider \[ D = \begin{bmatrix} 1 & x & y \\ 0 & 1 & z \\ 0 & 0 & 7 \end{bmatrix}. \] \(D\) has eigenvalues \(1, 1, 7\). If \(D\) is similar to \(A\), then \(D\) can also be diagonalized to \(\Lambda\) and hence must have two eigenvectors with eigenvalue 1. This is possible iff \(x = 0\). Our choice of \(C\) is obtained by letting \(y = z = 0\); our choice of \(N\) by letting \(y = 0, z = 1\). When \(x \neq 0\), \(D\) is not similar to \(A\); our choice of \(J\) is obtained by letting \(x = 1\). Note: there are choices for \(C\), \(N\), and \(J\) which are not upper triangular. (d) The matrix \(B_6 = A_6 - I\) has rank 1, so it has 5 independent eigenvectors with eigenvalue 0. It follows that the eigenvalues of \(B_6\) are \(0, 0, 0, 0, 0, 12 = tr(B_6)\), and the eigenvalues of \(A_6\) are \(1, 1, 1, 1, 1, 13\). We already know that \(A = S\Lambda S^{-1}\) for some \(S\) whose columns are independent eigenvectors of \(A\). But \(A\) is symmetric, so we can choose its eigenvectors to be orthonormal and have \(A = Q\Lambda Q^{-1}\), where \(Q\) is orthogonal.Keywords: positive definite, eigenvalues, similar matrices, rank one