4.4-final-2009¶
Final Exam, 2009 · Section 4.4: Orthonormal Bases and Gram-Schmidt Source: finalexam_1.pdf, p. 229
Problem¶
Problem 4: Consider the matrix
(a) Find an orthonormal basis for \(C(A)\) using Gram-Schmidt, forming the columns of a matrix \(Q\).
(b) Write each step of your Gram-Schmidt process from (a) as a multiplication of \(A\) on the __ (left or right) by some invertible matrix. Explain how the product of these invertible matrices relates to the matrix \(R\) from the QR factorization \(A = QR\) of \(A\).
(c) Gram-Schmidt on another matrix \(B\) (of the same size as \(A\)) gives the same orthonormal basis (the same \(Q\)) as in part (a). Which of the four subspaces, if any, must be the same for the matrices \(AA^T\) and \(BB^T\)? [You can do this part without doing (a) or (b).]
Solution
(18 points = 6 + 6 + 6) (a) From \(u_1 = (1, 1, 1, 1)^T\), we get \(q_1 = u_1/\|u_1\| = \frac{1}{2}(1, 1, 1, 1)^T\). \(v_2 = (1, -1, 0, 0)^T\), \(u_2 = v_2 - q_1^Tv_2q_1 = v_2 = (1, -1, 0, 0)^T\), \(q_2 = v_2/\|v_2\| = \frac{1}{\sqrt{2}}(1, -1, 0, 0)^T\); \(v_3 = (1, -1, -3, -1)^T\), \(u_3 = v_3 - q_1^Tv_3q_1 - q_2^Tv_3q_2 = v_3 + u_1 - u_2 = (1, 1, -2, 0)^T\), \(q_3 = u_3/\|u_3\| = \frac{1}{\sqrt{6}}(1, 1, -2, 0)^T\). Hence, we have \[ Q = \begin{bmatrix} 1/2 & 1/\sqrt{2} & 1/\sqrt{6} \\ 1/2 & -1/\sqrt{2} & 1/\sqrt{6} \\ 1/2 & 0 & -2/\sqrt{6} \\ 1/2 & 0 & 0 \end{bmatrix}. \] (b) Each step of the Gram-Schmidt process from (a) is a multiplication of \(A\) on the *right* as follows. \[ A \rightsquigarrow A \begin{bmatrix} 1/2 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix} \rightsquigarrow A \begin{bmatrix} 1/2 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1/\sqrt{2} & 0 \\ 0 & 0 & 1 \end{bmatrix} \] \[ \rightsquigarrow A \begin{bmatrix} 1/2 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1/\sqrt{2} & 0 \\ 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & 0 & 1 \\ 0 & 1 & -1 \\ 0 & 0 & 1 \end{bmatrix} \] \[ \rightsquigarrow A \begin{bmatrix} 1/2 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1/\sqrt{2} & 0 \\ 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & 0 & 1 \\ 0 & 1 & -1 \\ 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1/\sqrt{6} \end{bmatrix} = Q. \] The product of these invertible \(3 \times 3\) matrices is exactly \(R^{-1}\). (c) Since the Gram-Schmidt of \(A\) and \(B\) gives the same outcome, the column space of \(A\) and \(B\) are the same. We know that \(A\) and \(AA^T\) have the same column space, and \(B\) and \(BB^T\) have the same column space. Hence \(AA^T\) and \(BB^T\) have the same column space. Moreover, since the left nullspace is always orthogonal to the column space, \(AA^T\) and \(BB^T\) have the same left nullspace too. Also, notice that \(AA^T\) and \(BB^T\) are symmetric matrices, their row spaces are the same as the column spaces, and their nullspaces are the same as the left nullspaces. Therefore, all four subspaces of \(AA^T\) are the same as those of \(BB^T\).Keywords: Gram-Schmidt, QR factorization, orthonormal basis, column space