Skip to content

4.2-final-1998

Final Exam, 1998 · Section 4.2: Projections onto Lines and Subspaces Source: finalexam_1.pdf, pp. 7, 11

Problem

6. Suppose \(u_1, u_2, u_3\) is an orthonormal basis for \(R^3\) and \(v_1, v_2\) is an orthonormal basis for \(R^2\).

(a) (5 points) What is the rank, what are all vectors in the column space, and what is a basis for the nullspace for the matrix \(B = u_1(v_1 + v_2)^T\)?

(b) (5 points) Suppose \(A = u_1v_1^T + u_2v_2^T\). Multiply \(AA^T\) and simplify. Show that this is a projection matrix by checking the required properties.

(c) (4 points) Multiply \(A^T A\) and simplify. This is the identity matrix! Prove this (for example compute \(A^T Av_1\) and then finish the reasoning).

Solution (a) \(\text{rank}(B) = 1\), and all multiples of \(u_1\) are in the column space. The nullspace consists of the vectors \(x \in R^2\) with \((v_1 + v_2)^T x = 0\), so a basis for the nullspace is \(\{v_1 - v_2\}\). (The printed solution lists \(v_1 - v_2\) and \(v_3\), but \(B\) is \(3 \times 2\) so its nullspace is a subspace of \(R^2\) — \(v_3 \in R^3\) cannot be a nullspace vector, and rank-nullity gives \(\dim N(B) = 2 - 1 = 1\).) (b) \[ AA^T = (u_1v_1^T + u_2v_2^T)(v_1u_1^T + v_2u_2^T) = u_1u_1^T + u_2u_2^T \] since \(v_1^Tv_2 = 0\) (and \(v_1^Tv_1 = v_2^Tv_2 = 1\)). \(AA^T\) is symmetric and it equals \((AA^T)^2\): \[ (u_1u_1^T + u_2u_2^T)(u_1u_1^T + u_2u_2^T) = u_1u_1^T + u_2u_2^T. \] (The eigenvalues of \(AA^T\) are 1, 1, 0.) (c) \[ A^TA = v_1v_1^T + v_2v_2^T \quad \text{since } u_1^Tu_2 = 0. \] \[ A^TAv_1 = (v_1v_1^T + v_2v_2^T)v_1 = v_1, \qquad A^TAv_2 = (v_1v_1^T + v_2v_2^T)v_2 = v_2. \] Since \(v_1, v_2\) are a basis for \(R^2\), \(A^TAv = v\) for all \(v \in R^2\), i.e., \(A^TA = I\).

Keywords: orthonormal basis, projection matrix, column space, nullspace, rank