Skip to content

4.1-final-2012-strang

Final Exam, 2012 (strang) · Section 4.1: Orthogonality of Vectors and Subspaces Source: finalexam_2.pdf, pp. 81-82

Problem

5 (12 pts.) Suppose v is any unit vector in \(\mathbb{R}^3\). This question is about the matrix \(H\):

\[ H = I - 2vv^T. \]

(a) Multiply \(H\) times \(H\) to show that \(H^2 = I\).

(b) Show that \(H\) passes the tests for being a symmetric matrix and an orthogonal matrix.

(c) What are the eigenvalues of \(H\)? You have enough information to answer for any unit vector v, but you can choose one v and compute the \(\lambda\)'s.

Solution (a) \[ H^2 = (I - 2vv^T)^2 = I^2 - 4vv^T + 4(vv^T)^2 = I - 4vv^T + 4v(v^Tv)v^T = I - 4vv^T + 4vv^T = I. \] (b) Transpose is linear, \(I^T = I\), and anything of the form \(uu^T\) is symmetric: \[ (I - 2vv^T)^T = I - 2(v^T)^T v^T = I - 2vv^T. \] So \(H\) is symmetric. For orthogonality, use part (a) and symmetry: \[ HH^T = H^2 = I. \] (c) Since \(\|v\| = 1\), \[ Hv = v - 2(v^Tv)v = -v, \] so \(\lambda = -1\) is an eigenvalue (with eigenspace spanned by \(v\)). Let \(u \in (\mathrm{span}\{v\})^\perp\) be any vector orthogonal to \(v\). Then \[ Hu = u - 2(v^Tu)v = u, \] so \(\lambda = 1\) is also an eigenvalue. Since \((\mathrm{span}\{v\})^\perp\) is two-dimensional, we have found all eigenvalues: \[ \lambda = 1, 1, -1. \]

Keywords: Householder reflection, orthogonal matrix, symmetric matrix, eigenvalues