4.2-exam2-2011-edelman-b¶
Quiz 2, 2011 (edelman) · Section 4.2: Projections onto Lines and Subspaces Source: exam2_2.pdf, p. 45
Problem¶
4 (35 pts.)
-
(20 pts) The matrix \(A\) is \(2000 \times 2000\) and \(A^T A = I\). Let \(v\) be the vector \([1, 2, 3, \dots, 2000]^T\). Let \(v_1\) be the projection of \(v\) onto the space spanned by the first 1000 columns of \(A\). Let \(v_2\) be the projection of \(v\) onto the space spanned by the remaining 1000 columns of \(A\). What is \(v_1 + v_2\) in simplest form? Why? Give an example of a \(2000 \times 2000\) \(A\), where \(A^T A \neq I\), and where \(v_1 + v_2\) gives a different answer.
-
(15 pts) In a matrix \(A\), (which may not be invertible) the cofactors from the first row are \(C_{11}, C_{12}, \dots, C_{1n}\). Prove that the vector \(C = (C_{11}, C_{12}, \dots, C_{1n})\) is orthogonal to every row of \(A\) from row 2 to row \(n\). Hint: the dot product of \(C\) with row \(i\) (\(i = 2, \dots, n\)) is the determinant of what matrix?
Solution
1. \(v_1 + v_2 = v\), since projections onto orthogonal complements add to the identity. Here's something more explicit: let \(A\) take block form \([A_1 \ A_2]\). Then \(v_1\) and \(v_2\) are the projections onto the column spaces of \(A_1\) and \(A_2\) respectively. Since \(A^T A = I\) and \(A\) is square, \(A\) is orthogonal, so \(AA^T = I\), and we have \[ I = AA^T = A_1 A_1^T + A_2 A_2^T. \] Adding the two projections (recall in the orthogonal case this is just \(A_1 A_1^T v + A_2 A_2^T v\)) gives \(Iv = v\) by the above equation. An easy example where we get a different answer is if \(A\) is the zero matrix, where we have \(v_1 + v_2 = 0\) always for every \(v\). 2. Take the matrix \(A\) and replace row 1 with row \(i\). This matrix has two equal rows, hence 0 determinant. The determinant expansion by cofactors is exactly the desired dot product.Keywords: orthogonal matrix, projection, cofactors, determinant