Skip to content

4.4-exam2-2012

Quiz 2, 2012 · Section 4.4: Orthonormal Bases and Gram-Schmidt Source: exam2_2.pdf, pp. 21-22

Problem

3 (22 pts.)

The 3 × 3 matrix \(\begin{bmatrix} a & b & c \\ d & e & f \\ g & h & i \end{bmatrix}\) has \(QR\) decomposition

\[ \begin{bmatrix} a & b & c \\ d & e & f \\ g & h & i \end{bmatrix} = Q \begin{bmatrix} r_{11} & r_{12} & r_{13} \\ 0 & r_{22} & r_{23} \\ 0 & 0 & r_{33} \end{bmatrix}. \]

a) (7 pts.) What is \(r_{11}\) in terms of the variables \(a, b, c, d, e, f, g, h, i\)? (but not any of the elements of \(Q\).)

b) (15 pts.) Solve for \(x\) in the equation

\[ Q^Tx = \begin{bmatrix} 1 \\ 0 \\ 0 \end{bmatrix}, \]

expressing your answer possibly in terms of \(r_{11}, r_{22}, r_{33}\) and the variables \(a, b, c, d, e, f, g, h, i\), (but not any of the elements of \(Q\).)

Solution a) When we do a \(QR\) decomposition, we always start with the first column of our matrix, here the vector \((a \ d \ g)^T\), and normalize it to obtain the first column of \(Q\): \(q_1 = (a \ d \ g)^T / \|(a \ d \ g)^T\|\). Now, if we look at the first column of \(A = QR\), we have \[ (a \ d \ g)^T = r_{11}q_1 + 0 \cdot q_2 + 0 \cdot q_3 = r_{11}q_1 = r_{11} \frac{(a \ d \ g)^T}{\|(a \ d \ g)^T\|}, \] which implies that \(r_{11} = \|(a \ d \ g)^T\| = \sqrt{a^2 + d^2 + g^2}\). b) Look at the product \(Q^Tx = (1, 0, 0)^T\) row by row: we take the first row of \(Q^T\), that is, the first column of \(Q\), and take its dot product with \(x\) to obtain 1. We also take the second and third rows of \(Q^T\), that is, the second and third columns of \(Q\), and take their dot products with \(x\) to obtain 0. This means that \(x\) is perpendicular to the last 2 columns of \(Q\). But because \(Q\) has orthonormal columns and we are in \(R^3\), this can only mean that \(x\) is a multiple of the first column, say \(x = zq_1\) for some real number \(z\). But we said that \(q_1^Tx = 1\), which means \(q_1^T(zq_1) = zq_1^Tq_1 = 1\), and we know \(q_1^Tq_1 = 1\) because the columns of \(Q\) have norm 1. So clearly \(z = 1\) and \(x = q_1\), which we found in the previous question. So \[ x = q_1 = \frac{(a \ d \ g)^T}{\|(a \ d \ g)^T\|} = \frac{(a \ d \ g)^T}{r_{11}}. \]

Keywords: QR decomposition, orthonormal columns, Gram-Schmidt