Skip to content

6.2-exam3-2012

Quiz 3, 2012 · Section 6.2: Diagonalizing a Matrix Source: exam3_2.pdf, pp. 39-40

Problem

3. (28 pts.) We are told that \(A\) is \(2 \times 2\), symmetric, and Markov, and one of the real eigenvalues is \(y\) with \(-1 < y < 1\).

a) (7 pts.) What is this matrix \(A\) in terms of \(y\)?

b) (7 pts.) Compute the eigenvectors of \(A\).

c) (7 pts.) What is \(A^{2012}\) in simplest form?

d) (7 pts.) What is \(\lim_{n \to \infty} A^n\) in simplest form? (Explain briefly.)

Solution a) We have a symmetric matrix, hence \(A = \begin{bmatrix} a & b \\ b & c \end{bmatrix}\). Also, it is Markov, so we want \(a + b = 1\) and \(b + c = 1\), with all entries non-negative. So \(a = c\), and we have \(A = \begin{bmatrix} a & b \\ b & a \end{bmatrix}\). Now, we want the eigenvalues of this matrix to be \(y\) and 1 (recall that Markov matrices ALWAYS have 1 as an eigenvalue, with the all-ones vector as the corresponding eigenvector). The eigenvalues of \(A\) satisfy \(\det(A - \lambda I) = 0\), or \((a - \lambda)^2 - b^2 = 0\), or \(a - \lambda = \pm b\). So \(\lambda_1 = a + b = 1\) and \(\lambda_2 = a - b = y\) (since \(b \geq 0\)). Using \(a + b = 1\) in \(a - b = y\) we get \(2a - 1 = y\), or \(a = (y + 1)/2\), and then \(b = 1 - a = (1 - y)/2\). So we have found our symmetric Markov matrix with eigenvalues 1 and \(y\): \[ A = \begin{bmatrix} (1 + y)/2 & (1 - y)/2 \\ (1 - y)/2 & (1 + y)/2 \end{bmatrix}. \] b) An easy way to find the eigenvector corresponding to the eigenvalue 1 is to recall that we have a symmetric Markov matrix, so columns add to 1 but rows too, hence the constant vector will be an eigenvector. So for \(\lambda_1 = 1\) we have \(v_1 = (1/2, 1/2)^T\). And for \(\lambda_2 = y\), we find a vector in the nullspace of \[ A - yI = \begin{bmatrix} (1 - y)/2 & (1 - y)/2 \\ (1 - y)/2 & (1 - y)/2 \end{bmatrix}, \] which is easy: \(v_2 = (1/2, -1/2)^T\). c) We have now diagonalized \(A\): \(A = S\Lambda S^{-1}\), where the columns of \(S\) are the eigenvectors and \(\Lambda\) is a diagonal matrix with 1 and \(y\). So we have \[ A^{2012} = \begin{bmatrix} 1/2 & 1/2 \\ 1/2 & -1/2 \end{bmatrix}\begin{bmatrix} 1 & 0 \\ 0 & y \end{bmatrix}^{2012}\begin{bmatrix} 1 & 1 \\ 1 & -1 \end{bmatrix}, \] so that \[ A^{2012} = \begin{bmatrix} (1 + y^{2012})/2 & (1 - y^{2012})/2 \\ (1 - y^{2012})/2 & (1 + y^{2012})/2 \end{bmatrix}. \] d) From the above, and the fact that \(-1 < y < 1\), we can see clearly that \[ \lim_{n \to \infty} A^n = \begin{bmatrix} 1/2 & 1/2 \\ 1/2 & 1/2 \end{bmatrix}. \] Another way to reason: we know the steady state is the eigenvector of the dominating eigenvalue, in this case \(\lambda_1 = 1\), and so \(v_1 = (1/2, 1/2)^T\). But we are asking for the matrix which will give us this steady state, no matter what probability vector we start with, and so its column space has to be along the line of \(v_1\), and no bigger. But there is only one vector proportional to \(v_1\) which could also be a column of a Markov matrix, i.e. whose entries sum to 1. So both columns of the answer have to be \(v_1\). (You could also use the fact that the answer should be symmetric too.)

Keywords: Markov matrix, eigenvalues, eigenvectors, matrix powers, steady state