Eigenvalues of the product of one diagonal and one regular matrix

61 Views Asked by At

I'm reading a paper that makes the following claim that I don't understand.

Let $\Sigma, K\in\mathbb{R}^{n,n}$, where $\Sigma$ is a diagonal matrix with only non-negative real entries. Then the claim says that

$$\max_{i\in\{1,\cdots,n\}}Re(\lambda_i(K))\leq 0$$

implies

$$\max_{i\in\{1,\cdots,n\}}Re(\lambda_i(\Sigma K))\leq 0,$$

where $\lambda_i(\cdot)$ denotes the $i^{th}$ eigenvalue of the respective matrix. There are no other requirements to either $\Sigma$ or $K$ beyond the ones mentioned.

Does anybody know why this is true?

For reference, the paper claiming this is this paper in equations (3.6) to (3.8).

1

There are 1 best solutions below

3
On BEST ANSWER

As stated, the result is not true. For instance take $$ K=\begin{bmatrix} 2&-5\\ 5&-10\end{bmatrix},\ \ \ \Sigma=\begin{bmatrix}1&0\\0&0\end{bmatrix}. $$ The the eigenvalues of $K$ are negative, but $2$ is an eigenvalue of $\Sigma K$.

A small variation of the example, where $\Sigma$ is invertible, is to take
$$ \Sigma=\begin{bmatrix}1&0\\0&1/10\end{bmatrix}. $$ Then $\Sigma K$ has both eigenvalues with real part $+1/2$.