Skip to content

4.4-exam2-2007-johnson

Quiz 1, 2007 (johnson) · Section 4.4: Orthonormal Bases and Gram-Schmidt Source: exam2_1.pdf, pp. 153-154, 154-156

Problem

4 (28 pts.) Let

\[ A = \begin{bmatrix} 0 & 0 & 0 & 1 \\ 0 & 0 & 2 & 2 \\ 0 & 3 & 3 & 3 \\ 4 & 4 & 4 & 4 \end{bmatrix}. \]

Ordinary Gram-Schmidt would take the columns of \(A\) and produce an orthonormal basis \(Q\) with \(C(A) = C(Q)\). In this problem, we will modify that process and see what happens.

In particular, suppose that we proceed as in Gram-Schmidt, but we omit the normalization steps--we construct a basis of orthogonal vectors spanning \(C(A)\) but with lengths \(\ne 1\), by subtracting the projections as in Gram-Schmidt but skipping the division by the lengths. Let's call this "unnormalized Gram-Schmidt."

(a) Do "unnormalized Gram-Schmidt" on \(A\) to get an orthogonal but not orthonormal basis \(B\) for \(C(A)\).

(b) Compute the two (4-dimensional) volumes of the two parallelepipeds with edges given by the columns of \(A\) and the columns of \(B\).

Recall how ordinary Gram-Schmidt corresponded to multiplying \(A\) by a sequence of matrices, leading to the QR decomposition. Now, we want to look at unnormalized Gram-Schmidt in the same way, in order to see what it does to the volume (determinant). The next two parts refer to an arbitrary \(n \times n\) matrix \(A\) with independent columns, not the \(4 \times 4\) matrix from parts (a) and (b).

(c) For an arbitrary \(n \times n\) matrix \(A\) with (independent) columns \(a_1, a_2\), etcetera, write down the matrix \(M_2\) that you would multiply by \(A\) in the first step of unnormalized Gram-Schmidt to make the second column orthogonal to the first. What is \(\det M_2\)?

(d) Argue that the matrices \(M_3, M_4\), etcetera that you would multiply by in subsequent steps of unnormalized Gram-Schmidt all have the same determinant as \(M_2\). Therefore, the determinant of the final matrix \(B\) after unnormalized Gram-Schmidt is ______?

Solution 4a. Unnormalized Gram-Schmidt: \[ b_1 = a_1 = (0, 0, 0, 4)^T, \] \[ b_2 = a_2 - \frac{b_1^T a_2}{b_1^T b_1} b_1 = (0, 0, 3, 0)^T, \] \[ b_3 = a_3 - \frac{b_1^T a_3}{b_1^T b_1} b_1 - \frac{b_2^T a_3}{b_2^T b_2} b_2 = (0, 2, 0, 0)^T, \] \[ b_4 = a_4 - \frac{b_1^T a_4}{b_1^T b_1} b_1 - \frac{b_2^T a_4}{b_2^T b_2} b_2 - \frac{b_3^T a_4}{b_3^T b_3} b_3 = (1, 0, 0, 0)^T. \] Note that we must subtract off the projections onto the \(b\) vectors, not the \(a\) vectors--the \(b\) vectors span the same space, but are much simpler to project onto because they are orthogonal. Thus an orthogonal but not orthonormal basis \(B\) for \(C(A)\) is \[ B = \begin{bmatrix} 0 & 0 & 0 & 1 \\ 0 & 0 & 2 & 0 \\ 0 & 3 & 0 & 0 \\ 4 & 0 & 0 & 0 \end{bmatrix}. \] 4b. The volume of the parallelepiped with edges given by the columns of \(A\) is just the determinant (or rather, its absolute value, but here the determinant is positive anyway): \[ \det A = \det \begin{bmatrix} 0 & 0 & 0 & 1 \\ 0 & 0 & 2 & 2 \\ 0 & 3 & 3 & 3 \\ 4 & 4 & 4 & 4 \end{bmatrix} = 4 \cdot 3 \cdot 2 \cdot 1 = 24, \] where we have rearranged \(A\) into an upper-triangular matrix via two row swaps, and the determinant is then the product of the diagonals. The volume of the parallelepiped with edges given by the columns of \(B\) is similarly: \[ \det B = \det \begin{bmatrix} 0 & 0 & 0 & 1 \\ 0 & 0 & 2 & 0 \\ 0 & 3 & 0 & 0 \\ 4 & 0 & 0 & 0 \end{bmatrix} = (-1)^2 \cdot 4 \cdot 3 \cdot 2 \cdot 1 = 24 = \det A. \] 4c. We should have \[ [b_1 \; b_2 \; a_3 \; \cdots \; a_n] = [a_1 \; a_2 \; a_3 \; \cdots \; a_n] M_2. \] Notice that we must multiply \(M_2\) on the right since we are manipulating columns of \(A\). Since \(b_1 = a_1\), \(b_2 = a_2 - \frac{b_1^T a_2}{b_1^T b_1} b_1\), we must have \[ M_2 = \begin{bmatrix} 1 & -\frac{b_1^T a_2}{b_1^T b_1} & 0 & \cdots & 0 \\ 0 & 1 & 0 & \cdots & 0 \\ 0 & 0 & 1 & \cdots & 0 \\ \vdots & \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & 0 & \cdots & 1 \end{bmatrix}. \] Since this is an upper-triangular matrix with 1's on the diagonal, we have \(\det M_2 = 1\). 4d. In step \(k\), we only change the vector \(a_k\) to \(b_k\), which is \[ b_k = a_k - \text{linear combination of } b_1, \ldots, b_{k-1}, \] thus the matrix \(M_k\) is an upper-triangular matrix with diagonal entries 1 (and only nonzero off-diagonal entries in the \(k\)th column). Therefore, \(\det(M_k) = 1\) for all \(k\). This implies, finally, that \[ \det B = \det(A M_2 M_3 \cdots M_n) = \det A \cdot \det M_2 \cdots \det M_n = \det A. \]

Keywords: Gram-Schmidt, orthogonal basis, determinant, volume