Skip to content

6.3-exam3-2010

Quiz 3, 2010 · Section 6.3: Symmetric Positive Definite Matrices Source: exam3_2.pdf, pp. 2-3, 12

Problem

1. (30 points) In the following six problems produce a real \(2 \times 2\) matrix with the desired properties, or argue concisely, simply, and convincingly that no example can exist.

(a) (5 pts.) A \(2 \times 2\) symmetric, positive definite, Markov Matrix.

(b) (5 pts.) A \(2 \times 2\) symmetric, negative definite (i.e., negative eigenvalues), Markov Matrix.

(c) (5 pts.) A \(2 \times 2\) symmetric, Markov Matrix with one positive and one negative eigenvalue.

(d) (5 pts.) A \(2 \times 2\) matrix \(\neq 3I\) whose only eigenvalue is the double eigenvalue 3.

(e) (5 pts.) A \(2 \times 2\) symmetric matrix \(\neq 3I\) whose only eigenvalue is the double eigenvalue 3. (Note the word "symmetric" in problem (e).)

(f) (5 pts.) A \(2 \times 2\) non-symmetric matrix with eigenvalues 1 and \(-1\).

Solution (a) \(\begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix}\). (b) No example exists. There are many ways to see this. First way: negative definite means that all upper left submatrices have negative determinant. In particular, the \((1, 1)\) entry needs to be negative, but this violates the definition of Markov. Second way: since the trace is the sum of the eigenvalues, all negative eigenvalues implies that the trace is negative. But the diagonal entries are nonnegative by the Markov property, which is another violation. Third way: a Markov matrix always has 1 as an eigenvalue. All eigenvalues of a negative definite matrix are negative. (c) \(\begin{bmatrix} 0 & 1 \\ 1 & 0 \end{bmatrix}\). Really, any matrix of the form \(\begin{bmatrix} a & 1-a \\ 1-a & a \end{bmatrix}\) where \(1/2 > a \geq 0\) would do: then the trace would be \(< 1\), and since the trace is the sum of the eigenvalues and we know that one of the eigenvalues is 1, this means the other one has to be negative. (d) Take a Jordan block: \(\begin{bmatrix} 3 & 1 \\ 0 & 3 \end{bmatrix}\). (e) The answer we were looking for: all symmetric real matrices are diagonalizable. So if \(A\) is symmetric and has 3 as both of its eigenvalues, then its Jordan normal form is \(3I\), and \(A\) is similar to \(3I\). But the only matrix that is similar to \(3I\) is \(3I\) itself, which means \(A\) would have to be \(3I\). Hence no example exists. Another approach people tried but usually didn't fully justify: write the matrix as \(\begin{bmatrix} a & b \\ b & c \end{bmatrix}\). The trace is the sum of the eigenvalues and hence 6, and the determinant is the product of the eigenvalues and hence 9. So \(a + c = 6\) and \(ac - b^2 = 9\). Since \(b^2 \geq 0\), we get \(ac \geq 9\). The way to maximize \(ac\) subject to \(a + c = 6\) is to set \(a = c = 3\) (justified by elementary calculus or the arithmetic-mean/geometric-mean inequality). This means that \(a = c = 3\) and \(b = 0\), but then \(A = 3I\) again. (f) \(\begin{bmatrix} 1 & 1 \\ 0 & -1 \end{bmatrix}\).

Keywords: symmetric matrix, positive definite, Markov matrix, eigenvalues