4.4-exam2-2009¶
Quiz 2, 2009 · Section 4.4: Orthonormal Bases and Gram-Schmidt Source: exam2_1.pdf, pp. 166-167
Problem¶
Problem 2 (30 points = 6 + 10 + 10 + 4): The matrix
is converted to row-reduced echelon form by the usual row-elimination steps, resulting in the matrix
(\(\clubsuit\)) The minimum number of columns of \(A\) that form a dependent set of vectors is ___. The maximum number of columns of \(A\) that forms an independent set of vectors is ___.
(\(\diamondsuit\)) Give an orthonormal basis for the row space of \(A\). (Careful: be sure you start with a basis for the row space, not containing any dependent vectors.) Your answer may contain square roots left as some number.
(\(\spadesuit\)) Given the vector \(b = (2 \; 5 \; -9 \; 3)^T\), compute the closest vector \(p\) to \(b\) in the row space \(C(A^T)\). (Hint: less calculation is needed if you use your answer from (\(\diamondsuit\)).)
(\(\heartsuit\)) In terms of your answer \(p\) to (\(\spadesuit\)) above, what is the closest vector to \(b\) in the null space \(N(A)\)? (No calculation required, and you need not have solved (\(\spadesuit\)): you can leave your answer in terms of \(p\) and \(b\).)
Solution
Answers: (\(\clubsuit\)) 2, 2; (\(\diamondsuit\)) see below; (\(\spadesuit\)) \(p = (2 \; 4 \; 0 \; 4)^T\); (\(\heartsuit\)) \(b - p\). (\(\clubsuit\)) The key point is that the dependency of columns in \(R\) and \(A\) is the same. By inspection of \(R\) (or \(A\)), the first two columns are dependent, so that is the smallest dependent set. \(R\) has two pivots, so \(A\) has rank 2 and the column space is 2-dimensional; hence 2 is the maximum number of independent columns. Equivalently, the maximum number of independent columns is the number of columns in any basis for \(C(A)\), such as the 2 pivot columns. (\(\diamondsuit\)) The row-reduced echelon form \(R\) has the same row space as \(A\). We may therefore run Gram-Schmidt on the pivot rows of \(R\), which form a basis for the row space of \(R\) and \(A\): \[ q_1 = \frac{(1, 2, 0, 2)^T}{\|(1, 2, 0, 2)^T\|} = \frac{1}{3} (1, 2, 0, 2)^T. \] The second pivot row \((0, 0, 1, -9)^T\) has \(q_1^T a_2 = -6\), so \[ q_2 = \frac{(0, 0, 1, -9)^T + 6 q_1}{\|(0, 0, 1, -9)^T + 6 q_1\|} = \frac{(2, 4, 1, -5)^T}{\sqrt{46}}. \] REMARK: One can also obtain an orthonormal basis by starting with 2 rows of \(A\), since in this case any 2 rows are independent and form a basis. But the pivot rows of \(R\) are a nicer basis (more zeros), and the calculations are therefore much simpler. (\(\spadesuit\)) The closest vector \(p\) is the projection onto the row space: \[ p = (b^T q_1) q_1 + (b^T q_2) q_2 = 6 \cdot \frac{1}{3} (1, 2, 0, 2)^T + 0 = (2, 4, 0, 4)^T, \] since \(b^T q_1 = 6\) and \(b^T q_2 = 0\). (\(\heartsuit\)) The closest vector \(p\) to \(b\) in the row space is the projection onto the row space. The row space and the null space are orthogonal to each other, so \(b - p\) is exactly the orthogonal projection of \(b\) onto \(N(A)\); it is the closest vector to \(b\) in the null space.Keywords: row space, orthonormal basis, Gram-Schmidt, projection