4.4-exam2-2005¶
Quiz 2, 2005 · Section 4.4: Orthonormal Bases and Gram-Schmidt Source: exam2_1.pdf, pp. 107-108, 114-115
Problem¶
2 (30 pts.) This matrix \(Q\) has orthonormal columns \(q_1, q_2, q_3\):
(a) What equations must be satisfied by the numbers \(a, b, c, d\)? Is there a unique choice for those numbers, apart from multiplying them all by \(-1\)?
(b) Why is \(P = QQ^T\) a projection matrix? (Check the two properties of projections.) Why is \(QQ^T\) a singular matrix? Find the determinants of \(Q^T Q\) and \(QQ^T\).
(c) Suppose Gram-Schmidt starts with those same first two columns and with the third column \(a_3 = (1, 1, 1, 1)\). What third column would it choose for \(q_3\)? You may leave a square root not completed (if you want to).
Solution
2a. We need the vector \((a, b, c, d)\) to be a unit vector orthogonal to the other two columns of \(Q\). Hence the numbers \(a, b, c, d\) must satisfy the following equations: \[ a^2 + b^2 + c^2 + d^2 = 1, \qquad .1 a + .7 b + .1 c + .7 d = 0, \qquad .5 a + .5 b - .5 c - .5 d = 0. \] The last two equations define a two-dimensional subspace of \(\mathbb{R}^4\). Taking two independent vectors in this subspace and normalizing them gives two good choices of \(a, b, c, d\) that are not merely negatives of each other. Hence the choice of the last column of \(Q\) is not unique: no. 2b. \(P\) is a projection matrix if it satisfies two conditions: \(P\) is symmetric and \(P^2 = P\). Indeed, \[ P^T = (QQ^T)^T = (Q^T)^T Q^T = QQ^T = P, \qquad P^2 = Q(Q^T Q) Q^T = QQ^T = P, \] since \(Q^T Q = I\) by orthonormality of the columns of \(Q\). (Another valid argument is that \(P\) is the matrix for projection onto the column space of \(Q\), as can be easily checked using the projection matrix formula.) The matrix \(P = QQ^T\) is singular because both \(Q\) and \(Q^T\) have rank 3, and hence their product \(P\) has rank at most 3 (in fact, exactly 3). Since \(P\) is a 4 by 4 matrix, it is not full rank and hence singular. Since \(QQ^T\) is singular, \(\det QQ^T = 0\). Also, \(\det Q^T Q = \det I = 1\). 2c. The Gram-Schmidt process first produces a vector \(Q_3\) orthogonal to both \(q_1\) and \(q_2\): \[ Q_3 = a_3 - (q_1 \cdot a_3) q_1 - (q_2 \cdot a_3) q_2 \] (remember that \(q_1\) and \(q_2\) are unit vectors). We have \(q_1 \cdot a_3 = 1.6\) and \(q_2 \cdot a_3 = 0\), so \[ Q_3 = (1, 1, 1, 1) - 1.6 (.1, .7, .1, .7) = (.84, -.12, .84, -.12). \] Normalizing, we get \[ q_3 = \frac{Q_3}{\|Q_3\|} = (.7, -.1, .7, -.1). \]Keywords: orthonormal columns, projection matrix, determinant, Gram-Schmidt