6.2-exam3-2008-strang-b¶
Quiz 3, 2008 (strang) · Section 6.2: Diagonalizing a Matrix Source: exam3_1.pdf, pp. 148-149
Problem¶
3 (30 pts.) Suppose the vectors \(x, y\) give an orthonormal basis for \(R^2\) and \(A = xy^T\).
(a) Compute the rank of \(A\) and the rank of \(A^2 = (xy^T)(xy^T)\). Use this information to find the eigenvalues of \(A\).
(b) Explain why this matrix \(B\) is similar to \(A\) (and write down what similar means):
\[
B = \left[ \begin{array}{c} x^T \\ y^T \end{array} \right] A \left[ \begin{array}{cc} x & y \end{array} \right]
\]
(c) The eigenvalues of \(Q\) are \(\lambda_1 = e^{i\theta} = \cos \theta + i \sin \theta\) and \(\lambda_2 = e^{-i\theta} = \cos \theta - i \sin \theta\):
\[
\text{Rotation matrix } Q = \left[ \begin{array}{cc} \cos \theta & -\sin \theta \\ \sin \theta & \cos \theta \end{array} \right]
\]
Find the eigenvectors of \(Q\). Are they perpendicular?
Solution
(a) Any matrix given by \(A = xy^T\) for two non-zero vectors \(x, y\) will have rank 1. Every row will be a multiple of \(y^T\), and every column will be a multiple of \(x\), meaning that it must have rank 1. Alternatively, we note that \(x\) is in the nullspace of \(A\) (since \(y^Tx = 0\)), and that \(y\) is not (since \(y^Ty = 1\)), so that \(A\) must have rank exactly 1. Note that \[ A^2 = (xy^T)(xy^T) = x(y^Tx)y^T \] is the zero matrix, since \(y^Tx = 0\) (they are perpendicular vectors). So \(A^2\) has rank 0. If the eigenvalues of \(A\) are \(\lambda_1\) and \(\lambda_2\), then the eigenvalues of \(A^2\) are \(\lambda_1^2\) and \(\lambda_2^2\). Since \(A^2\) only has the eigenvalue 0, both \(\lambda_1\) and \(\lambda_2\) must be 0. (b) Two square matrices \(A\) and \(B\) are similar if there is some invertible matrix \(M\) such that \(B = MAM^{-1}\). Similarity is *not* the same thing as having equal eigenvalues; this only works if both \(A\) and \(B\) are diagonalizable matrices, and in fact our \(A\) is not diagonalizable. To be more precise, similarity implies that \(A\) and \(B\) have equal eigenvalues, but the converse is not true. In this case we check that \(A\) and \(B\) are similar by showing that the other factors are inverses: \[ \begin{bmatrix} x^T \\ y^T \end{bmatrix} \begin{bmatrix} x & y \end{bmatrix} = \begin{bmatrix} x^Tx & x^Ty \\ y^Tx & y^Ty \end{bmatrix} = \begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix}, \] since \(x\) and \(y\) are perpendicular, and both of unit length. (c) Given the eigenvalues of \(Q\), we find the eigenvectors using \(N(Q - \lambda I)\). We start with \(\lambda_1 = \cos \theta + i \sin \theta\): \[ Q - (\cos \theta + i \sin \theta)I = \begin{bmatrix} -i \sin \theta & -\sin \theta \\ \sin \theta & -i \sin \theta \end{bmatrix}, \] and this matrix has nullspace generated by \((1, -i)\) or equivalently \((i, 1)\). Similarly, for \(\lambda_2 = \cos \theta - i \sin \theta\) we find \[ Q - (\cos \theta - i \sin \theta)I = \begin{bmatrix} i \sin \theta & -\sin \theta \\ \sin \theta & i \sin \theta \end{bmatrix}, \] which has nullspace generated by \((1, i)\). Every orthogonal matrix has perpendicular eigenvectors. We check in this specific case: \[ \begin{bmatrix} i \\ 1 \end{bmatrix} \cdot \begin{bmatrix} 1 \\ i \end{bmatrix} = \begin{bmatrix} i \\ 1 \end{bmatrix}^H \begin{bmatrix} 1 \\ i \end{bmatrix} = \begin{bmatrix} -i & 1 \end{bmatrix} \begin{bmatrix} 1 \\ i \end{bmatrix} = 0. \] Make sure to take \(x_1^H x_2\) and not \(x_1^T x_2\).Keywords: rank, eigenvalues, similar matrices, rotation matrix, complex eigenvectors