6.3-final-2012-strang¶
Final Exam, 2012 (strang) · Section 6.3: Symmetric Positive Definite Matrices Source: finalexam_2.pdf, pp. 77-78
Problem¶
3 (12 pts.)
(a) Find the three pivots and the determinant of \(A\):
\[
A = \begin{bmatrix} 1 & 0 & -1 \\ 0 & 1 & 1 \\ -1 & 1 & 0 \end{bmatrix}.
\]
(b) The rank of \(A - I\) is _, so that \(\lambda =\) _ is an eigenvalue. The remaining two eigenvalues of \(A\) are \(\lambda =\) __. These eigenvalues are all ______, because \(A^T = A\).
(c) The unit eigenvectors \(x_1, x_2, x_3\) will be orthonormal. Prove that
\[
A = \lambda_1x_1x_1^T + \lambda_2x_2x_2^T + \lambda_3x_3x_3^T.
\]
You may compute the \(x_i\)'s and use numbers. Or, without numbers, you may show that the right side has the correct eigenvectors \(x_1, x_2, x_3\) with eigenvalues \(\lambda_1, \lambda_2, \lambda_3\).
Solution
(a) Elimination gives \[ A \sim \begin{bmatrix} 1 & 0 & -1 \\ 0 & 1 & 1 \\ 0 & 0 & -2 \end{bmatrix}. \] Thus the pivots are \(1, 1, -2\). Since we reduced \(A\) without any row switches (permutations) or row scalings, \[ \det A = 1 \cdot 1 \cdot (-2) = -2. \] (b) \(\mathrm{rank}(A - I) = 2\), so \(\dim N(A - I) = 1\), and \[ \lambda = 1 \] is an eigenvalue of algebraic and geometric multiplicity one. The remaining two eigenvalues of \(A\) are \[ \lambda = -1, \ 2. \] The eigenvalues are all real, because \(A\) is symmetric. (c) As suggested, check that the right side does the correct thing on the basis \(\{x_1, x_2, x_3\}\): \[ (\lambda_1x_1x_1^T + \lambda_2x_2x_2^T + \lambda_3x_3x_3^T) x_i = \lambda_i (x_i^T x_i) x_i = \lambda_i x_i = A x_i. \] Having checked this on a basis, by linearity of matrix multiplication the two expressions agree on every vector, so the matrices are identical. For the record, the three unit eigenvectors are \[ x_1 = \frac{1}{\sqrt{2}} \begin{bmatrix} 1 \\ 1 \\ 0 \end{bmatrix}, \qquad x_2 = \frac{1}{\sqrt{6}} \begin{bmatrix} 1 \\ -1 \\ 2 \end{bmatrix}, \qquad x_3 = \frac{1}{\sqrt{3}} \begin{bmatrix} 1 \\ -1 \\ -1 \end{bmatrix}. \]Keywords: pivots, determinant, eigenvalues, spectral theorem