Skip to content

4.4-final-2008-strang-b

Final Exam, 2008 (strang) · Section 4.4: Orthonormal Bases and Gram-Schmidt Source: finalexam_1.pdf, p. 216

Problem

6 (12 pts.) (a) Suppose \(q_1, q_2, a_3\) are linearly independent, and \(q_1\) and \(q_2\) are already orthonormal. Give a formula for a third orthonormal vector \(q_3\) as a linear combination of \(q_1, q_2, a_3\).

(b) Find the vector \(q_3\) in part (a) when

\[ q_1 = \frac{1}{2} \begin{bmatrix} 1 \\ 1 \\ 1 \\ 1 \end{bmatrix} \quad q_2 = \frac{1}{2} \begin{bmatrix} 1 \\ -1 \\ 1 \\ -1 \end{bmatrix} \quad a_3 = \begin{bmatrix} 1 \\ 2 \\ 3 \\ 4 \end{bmatrix} \]

(c) Find the projection matrix \(P\) onto the subspace spanned by the first two vectors \(q_1\) and \(q_2\). You can give a formula for \(P\) using \(q_1\) and \(q_2\) or give a numerical answer.

Solution (a) This is the Gram-Schmidt process. We define \[ w_3 = a_3 - (q_1 \cdot a_3)q_1 - (q_2 \cdot a_3)q_2 \] and then set \(q_3 = w_3/\|w_3\|\). Note that we do not need denominators in the expression for \(w_3\) because the \(q_i\) are already unit vectors. (b) Substituting in, we find \[ w_3 = a_3 - 5q_1 - (-1)q_2 = (-1, -1, 1, 1). \] Renormalizing we get \(q_3 = \frac{1}{2}(-1, -1, 1, 1)\). (c) The projection matrix \(P\) is exactly the expression we used for Gram-Schmidt: \(P = q_1q_1^T + q_2q_2^T\). There are other more complicated expressions which are also correct. We can start at the most general and simplify to get this one; if \(A\) has columns \(q_1\) and \(q_2\) then \[ P = A(A^TA)^{-1}A^T = A(I)A^T = q_1q_1^T + q_2q_2^T, \] where we used the column-row picture of multiplication for the last step.

Keywords: Gram-Schmidt, orthonormal vectors, projection matrix