6.2-final-2014-strang¶
Final Exam, 2014 · Section 6.2: Diagonalizing a Matrix Source: finalexam_2.pdf, p. 139
Problem¶
5 (15 points)
(a) Suppose an \(n\) by \(n\) matrix \(A\) has \(n\) independent eigenvectors \(x_1, \dots, x_n\) with eigenvalues \(\lambda_1, \dots, \lambda_n\). What matrix equation would you solve for \(c_1, \dots, c_n\) to write the vector \(u_0\) as a combination \(u_0 = c_1x_1 + \dots + c_nx_n\)?
(b) Suppose a sequence of vectors \(u_0, u_1, u_2, \dots\) starts from \(u_0\) and satisfies \(u_{k+1} = Au_k\). Find the vector \(u_k\) as a combination of \(x_1, \dots, x_n\).
(c) State the exact requirement on the eigenvalues \(\lambda\) so that \(A^k u_0 \to 0\) as \(k \to \infty\) for every vector \(u_0\). Prove that your condition must hold.
Solution
(a) We solve the linear system \[ [x_1 \ x_2 \dots x_n] \begin{bmatrix} c_1 \\ \vdots \\ c_n \end{bmatrix} = u_0, \] i.e. \(Sc = u_0\) where \(S\) is the eigenvector matrix. The \(x_i\)'s are independent, so \(S\) is invertible and \(c = S^{-1}u_0\) is unique. (b) Applying \(A\) to each eigenvector multiplies it by its eigenvalue: \[ u_k = A^k u_0 = \sum_{i=1}^{n} c_i \lambda_i^k x_i. \] (c) The exact requirement is \[ |\lambda_i| < 1 \quad \text{for every } i. \] If this holds, then every coefficient \(c_i \lambda_i^k \to 0\), so \(u_k \to 0\). For the converse, since the statement must hold for every \(u_0\), we may take \(u_0 = x_i\) (so \(c_i = 1\)); then \(u_k = \lambda_i^k x_i\) must tend to 0, which forces \(|\lambda_i| < 1\) for each \(i\).Keywords: powers of A, eigenvalues, eigenvector expansion, stability