Skip to content

6.2-exam3-2006-b

Quiz 3, 2006 · Section 6.2: Diagonalizing a Matrix Source: exam3_1.pdf, pp. 104, 111-112

Problem

3 (33 pts.) Suppose the \(n\) by \(n\) matrix \(A\) has \(n\) orthonormal eigenvectors \(q_1, \dots, q_n\) and \(n\) positive eigenvalues \(\lambda_1, \dots, \lambda_n\). Thus \(Aq_j = \lambda_j q_j\).

(a) What are the eigenvalues and eigenvectors of \(A^{-1}\)? Prove that your answer is correct.

(b) Any vector \(b\) is a combination of the eigenvectors:

\[ b = c_1q_1 + c_2q_2 + \dots + c_nq_n. \]

What is a quick formula for \(c_1\) using orthogonality of the \(q\)'s?

(c) The solution to \(Ax = b\) is also a combination of the eigenvectors:

\[ A^{-1}b = d_1q_1 + d_2q_2 + \dots + d_nq_n. \]

What is a quick formula for \(d_1\)? You can use the \(c\)'s even if you didn't answer part (b).

Solution (a) \(A^{-1}\) has eigenvalues \(\frac{1}{\lambda_j}\) with the same eigenvectors: \[ Aq_j = \lambda_j q_j \longrightarrow q_j = \lambda_j A^{-1}q_j \longrightarrow A^{-1}q_j = \frac{1}{\lambda_j}q_j. \] (b) Multiply \(b = c_1q_1 + \dots + c_nq_n\) by \(q_1^{T}\). Orthogonality gives \[ q_1^{T}b = c_1q_1^{T}q_1, \quad \text{so} \quad c_1 = \frac{q_1^{T}b}{q_1^{T}q_1} = q_1^{T}b, \] since the \(q\)'s are unit vectors. (c) Multiplying \(b\) by \(A^{-1}\) multiplies each \(q_i\) by \(\frac{1}{\lambda_i}\) (part (a)), so \(c_i\) becomes \[ d_1 = \frac{c_1}{\lambda_1} \left( = \frac{q_1^{T}b}{\lambda_1 q_1^{T}q_1} = \frac{q_1^{T}b}{\lambda_1} \right). \]

Keywords: eigenvalues, eigenvectors, orthonormal basis, inverse matrix