6.2-exam3-2012-strang¶
Quiz 3, 2012 (strang) · Section 6.2: Diagonalizing a Matrix Source: exam3_2.pdf, pp. 46-48
Problem¶
2. (30 pts.) This question is about the real matrix
\[
A = \begin{bmatrix} 1 & c \\ 1 & -1 \end{bmatrix}, \quad \text{for } c \in \mathbb{R}.
\]
(a) Find the eigenvalues of \(A\), depending on \(c\). For which values of \(c\) does \(A\) have real eigenvalues?
(b) For one particular value of \(c\), convince me that \(A\) is similar to both the matrix
\[
B = \begin{bmatrix} 2 & 0 \\ 0 & -2 \end{bmatrix},
\]
and to the matrix
\[
C = \begin{bmatrix} 2 & 2 \\ 0 & -2 \end{bmatrix}.
\]
Don't forget to say which value \(c\) this happens for.
(c) For one particular value of \(c\), convince me that \(A\) cannot be diagonalized. It is not similar to a diagonal matrix \(\Lambda\), when \(c\) has that value. Which value \(c\)? Why not?
Solution
(a) Since \(\operatorname{tr}(A) = 0 = \lambda_1 + \lambda_2\), we see that \(\lambda_2 = -\lambda_1\). Also, \[ -1 - c = \det(A) = \lambda_1\lambda_2 = -\lambda_1^2. \] Thus \(\lambda_1^2 = 1 + c\), and \[ \lambda = \pm\sqrt{1 + c}. \] Therefore, the eigenvalues are real precisely when \(c \geq -1\). (b) If two matrices are similar, then they have the same eigenvalues (those are \(2, -2\) for both \(B\) and \(C\)). Here we must therefore have \(0 = \operatorname{tr}(A)\) and \(-1 - c = \det(A) = -4\). We see that this happens precisely when \(c = 3\), where we check that indeed the eigenvalues are \(2, -2\). However, this does not guarantee that they are similar -- and hence is not convincing. Convincing: The eigenvalues \(2, -2\) are different, so \(A\), \(B\), and \(C\) are all diagonalizable, with the same diagonal matrix (for example \(\Lambda = B\)!). Therefore \(A\), \(B\), and \(C\) are all similar when \(c = 3\). (c) As we saw above, \(\operatorname{tr}(A) = 0\), so regardless of \(c\) the eigenvalues come in pairs \(\lambda_2 = -\lambda_1\). This means that whenever \(\lambda_1 \neq 0\), we have two different eigenvalues, and hence \(A\) is diagonalizable (not what we're after). Thus we need \(\lambda_1 = \lambda_2 = 0\), a repeated eigenvalue, which happens when \(c = -1\) (so \(\det A = 0\)) as the only suspect -- does it work? Convincing: For \(c = -1\), we have \[ N(A - 0 \cdot I) = \operatorname{span}\left\{ \begin{bmatrix} 1 \\ 1 \end{bmatrix} \right\}. \] With only a 1-dimensional space of eigenvectors for the matrix, we are convinced that \(A\) is not diagonalizable for \(c = -1\).Keywords: eigenvalues, similar matrices, diagonalizable, trace, determinant