4.4-exam2-2000¶
Quiz 2, 2000 · Section 4.4: Orthonormal Bases and Gram-Schmidt Source: exam2_1.pdf, pp. 41, 44
Problem¶
1 (36 pts.) Suppose \(Q\) is a 4 by 3 matrix with orthonormal columns \(q_1, q_2, q_3\).
(a) Starting from a vector \(v\) (not in the column space of \(Q\)), give a formula for the fourth orthonormal vector \(q_4\) that is produced by Gram-Schmidt from \(q_1, q_2, q_3, v\).
(b) Describe the nullspace of \(Q\) (the same 4 by 3 matrix) and the nullspace of \(Q^T\). (You can answer even if you didn't find the particular formula for \(q_4\) in part a.) Describe also the nullspaces of \(Q^T Q\) and \(QQ^T\).
(c) Suppose \(b = q_1 + 2q_2 + 3q_3 + 4q_4\). Find the least squares solution \(\bar{x}\) to \(Qx = b\). What is the projection \(p\) of this \(b\) onto the column space of \(Q\)?
Solution
1a. \[ q_4^* = v - (q_1^T v)q_1 - (q_2^T v)q_2 - (q_3^T v)q_3, \qquad q_4 = \frac{q_4^*}{\|q_4^*\|}. \] 1b. The nullspace of \(Q\) is just the zero vector (\(Q\) has a pivot in every column). The nullspace of \(Q^T\) has dimension one and consists of all scalar multiples of \(q_4\) (because we know \(q_4\) is orthogonal to \(q_1, q_2\) and \(q_3\)). The nullspace of \(Q^T Q = I\) is just the zero vector. The nullspace of \(QQ^T\) again has dimension one and is all scalar multiples of \(q_4\). 1c. \(Q^T Q \bar{x} = Q^T b\) is the same as \(\bar{x} = Q^T b\), so \[ \bar{x} = \begin{bmatrix} q_1^T(q_1 + 2q_2 + 3q_3 + 4q_4) \\ q_2^T(q_1 + 2q_2 + 3q_3 + 4q_4) \\ q_3^T(q_1 + 2q_2 + 3q_3 + 4q_4) \end{bmatrix} = \begin{bmatrix} 1 \\ 2 \\ 3 \end{bmatrix}. \] The projection is \(p = Q\bar{x} = q_1 + 2q_2 + 3q_3\).Keywords: orthonormal columns, Gram-Schmidt, nullspace, least squares, projection