7.1-final-2015¶
Final Exam, 2015 · Section 7.1: Singular Values and Singular Vectors Source: finalexam_2.pdf, p. 155
Problem¶
3 (5+5+5 pts.) Suppose I give you an orthonormal basis \(q_1, \dots, q_4\) for \(\mathbb{R}^4\) and an orthonormal basis \(z_1, \dots, z_6\) for \(\mathbb{R}^6\). From these you create the 6 by 4 matrix \(A = z_1q_1^T + z_2q_2^T\).
(a) Find a basis for the nullspace of \(A\).
(b) Find a particular solution to \(Ax = z_1\) and find the complete solution.
(c) Find \(A^TA\) and find an eigenvector of \(A^TA\) with \(\lambda = 1\).
Solution
(a) The matrix has SVD \(A = Z J Q^T\), where \(J\) is the \(6 \times 4\) diagonal matrix with diagonal entries \((1, 1, 0, 0)\) (columns \(z_1, z_2, z_3, z_4\) receive the images \(q_1, q_2, 0, 0\)). The nullspace consists of the columns of \(Q\) corresponding to zero singular values: \[ N(A) = \mathrm{span}\{q_3, q_4\}. \] (b) One particular solution is \(x = q_1\), since \[ (z_1q_1^T)q_1 = z_1(q_1^Tq_1) = z_1, \qquad (z_2q_2^T)q_1 = z_2(q_2^Tq_1) = 0, \] by orthonormality. The complete solution adds the nullspace: \[ x = q_1 + c\, q_3 + d\, q_4 \] for arbitrary scalars \(c, d\). (c) Expanding, \[ A^T A = (q_1z_1^T + q_2z_2^T)(z_1q_1^T + z_2q_2^T) = (z_1 \cdot z_1) q_1q_1^T + (z_1 \cdot z_2) q_1q_2^T + (z_2 \cdot z_1) q_2q_1^T + (z_2 \cdot z_2) q_2q_2^T. \] The cross terms vanish because \(z_1 \cdot z_2 = 0\), and \(z_1 \cdot z_1 = z_2 \cdot z_2 = 1\), so \[ A^T A = q_1q_1^T + q_2q_2^T. \] Then \(A^TAq_1 = q_1\) and \(A^TAq_2 = q_2\), so \(q_1\) (and \(q_2\), or any nonzero combination of them) is an eigenvector of \(A^TA\) with \[ \lambda = 1. \]Keywords: SVD, nullspace, particular solution, orthonormal basis