4.2-exam2-2011¶
Quiz 2, 2011 · Section 4.2: Projections onto Lines and Subspaces Source: exam2_2.pdf, pp. 8-9
Problem¶
- (8 pts. each) Suppose \(a_1\) and \(a_2\) are orthogonal unit vectors in \(R^5\).
(a) What are the requirements on a matrix \(P\) to be a projection matrix? Verify that \(P = a_1a_1^T + a_2a_2^T\) satisfies those requirements.
(b) If \(a_3\) is in \(R^5\), what combination of \(a_1\) and \(a_2\) is closest to \(a_3\)?
(c) Find a combination \(c\) of \(a_1\), \(a_2\), \(a_3\) that is perpendicular to \(a_1\) and \(a_2\). If possible, choose \(c \neq 0\). Describe all cases when \(c = 0\) is the only possibility.
(d) Show that \(a_1\) and \(a_2\) and \(c\) are eigenvectors of \(P\) (if \(c \neq 0\)) and find their eigenvalues.
Solution
(a) \(P\) is a projection (orthogonal projection) if \[ P^2 = P \quad \text{and} \quad P^T = P. \] We check this for \(P = a_1a_1^T + a_2a_2^T\). Since \(a_1^Ta_2 = 0\), \(a_2^Ta_1 = 0\), and \(a_1^Ta_1 = a_2^Ta_2 = 1\), we get \[ P^2 = (a_1a_1^T + a_2a_2^T)(a_1a_1^T + a_2a_2^T) = a_1a_1^Ta_1a_1^T + a_1a_1^Ta_2a_2^T + a_2a_2^Ta_1a_1^T + a_2a_2^Ta_2a_2^T = a_1a_1^T + a_2a_2^T = P, \] and \[ P^T = (a_1a_1^T + a_2a_2^T)^T = (a_1^T)^Ta_1^T + (a_2^T)^Ta_2^T = a_1a_1^T + a_2a_2^T = P. \] (b) The closest combination is \(Pa_3 = (a_1^Ta_3)a_1 + (a_2^Ta_3)a_2\). (c) \(c = a_3 - Pa_3 = a_3 - (a_1^Ta_3)a_1 - (a_2^Ta_3)a_2\). \(c = 0\) only if \(a_3\) is in the plane generated by \(a_1\) and \(a_2\). (d) Since \(P\) is the projection onto the column space of \(A = [a_1 \mid a_2]\), we have \[ Pa_1 = a_1 \implies \lambda_1 = 1, \qquad Pa_2 = a_2 \implies \lambda_2 = 1, \qquad Pc = 0 \implies \lambda_3 = 0. \]Keywords: projection matrix, orthonormal vectors, eigenvectors, eigenvalues