6.3-exam3-2014-strang¶
Exam III, 2014 (strang) · Section 6.3: Symmetric Positive Definite Matrices Source: exam3_2.pdf, pp. 91, 97
Problem¶
2. (a) (24 points) Suppose a symmetric \(n\) by \(n\) matrix \(S\) has eigenvalues \(\lambda_1 > \lambda_2 > \dots > \lambda_n\) and orthonormal eigenvectors \(q_1, \dots, q_n\). If \(x = c_1q_1 + c_2q_2 + \dots + c_nq_n\) show that \(x^Tx = c_1^2 + \dots + c_n^2\) and \(x^TSx = \lambda_1c_1^2 + \dots + \lambda_nc_n^2\).
(b) What is the largest possible value of \(R(x) = \frac{x^TSx}{x^Tx}\) for nonzero \(x\)? Describe a vector \(x\) that gives this maximum value for this ratio \(R(x)\)?
Solution
(a) Since the eigenvectors are orthonormal, one has \[ x^Tx = (c_1q_1 + \dots + c_nq_n)^T(c_1q_1 + \dots + c_nq_n) = c_1^2q_1^Tq_1 + \dots + c_n^2q_n^Tq_n = c_1^2 + \dots + c_n^2. \] On the other hand, \[ x^TSx = (c_1q_1 + \dots + c_nq_n)^TS(c_1q_1 + \dots + c_nq_n) = (c_1q_1 + \dots + c_nq_n)^T(\lambda_1c_1q_1 + \dots + \lambda_nc_nq_n) = \lambda_1c_1^2 + \dots + \lambda_nc_n^2. \] (b) Using (a), \[ R(x) = \frac{\lambda_1c_1^2 + \dots + \lambda_nc_n^2}{c_1^2 + \dots + c_n^2}. \] Since \(\lambda_1 > \lambda_2 > \dots > \lambda_n\), \(R(x)\) is maximal when \(c_2 = \dots = c_n = 0\) and \(c_1 \neq 0\). In this case the largest value of \(R(x)\) is \(\lambda_1\), and the associated vector \(x\) is any nonzero multiple of \(q_1\).Keywords: symmetric matrix, eigenvalues, orthonormal eigenvectors, Rayleigh quotient