The matrices that I am considering are

$$ M = \begin{bmatrix} A & B \\ C & -A^\top \end{bmatrix}, $$

with $A,B,C\in\mathbb{C}^{n\times n}$, $C = C^\top$ and $B = B^\top$. I noticed from numerical calculations that the eigenvalues are symmetric with respect to the origin, i.e., if $\lambda$ is an eigenvalue of $M$ then $-\lambda$ is as well.

I have not been able to shown that this has to be the case. Initially I tried to see if I could come up with a similarity transformation which would make it block diagonal, with one block the negative of the other, but with no success. Next I tried to use some tricks to manipulate the determinant, namely I showed that

\begin{align} \det(M - \lambda\,I) &= \det(B) \det((-A^\top - \lambda\,I) B^{-1} (A - \lambda\,I) - C) \\ &= \det(C) \det((A - \lambda\,I) C^{-1} (-A^\top - \lambda\,I) - B) \\ \end{align}

assuming that either $\det(B)\neq0$ or $\det(C)\neq0$. However, this did not seem to bring me any closer to showing that all eigenvalues are mirrored around the imaginary axis.

  • 1
    $\begingroup$ Are you sure you want to take transpose and not conjugated transpose? Also, the reflection of $\lambda$ to the imaginary axis is not $-\lambda$ but $-\bar\lambda$. $\endgroup$
    – Berci
    Feb 8 '19 at 1:09
  • 1
    $\begingroup$ I have constructed a $8\times8$ matrix with the form of your matrix $M$ in Octave, and the eigenvalues are not mirrored around the imaginary axis, but if $\lambda$ is an eigenvalue, then $-\lambda$ is an eigenvalue as well. $\endgroup$
    – jobe
    Feb 8 '19 at 1:15
  • $\begingroup$ @Berci I did mean the normal transpose, but initially I did some test calculations using real matrices, which also have complex conjugate eigenvalues, so did not notice that it was actually a origin reflection instead of axis reflection. I also just figured out that my definition of $M$ is essentially the definition of an Hamiltonian matrix. $\endgroup$ Feb 8 '19 at 1:15
  • $\begingroup$ @KwinvanderVeen: I'd like to see how that works! $\endgroup$ Feb 8 '19 at 1:25

The definition of $M$ is the same as that of an Hamiltonian matrix. In order to show that if $\lambda$ is an eigenvalue then $-\lambda$ is an eigenvalue as well, the following identity can be used

$$ \det(M - \lambda\,I) = \det(M^\top - \lambda\,I) $$

Defining a new matrix as

$$ J = \begin{bmatrix} 0 & -I \\ I & 0 \end{bmatrix}, $$

where it can be noted that $\det(J)=-1$ and $J^{-1} = J^\top = -J$. Using this matrix it can be shown that $J\,M^\top J = M$, namely

\begin{align} J\,M^\top J &= \begin{bmatrix} 0 & -I \\ I & 0 \end{bmatrix} \begin{bmatrix} A & B \\ C & -A^\top \end{bmatrix}^\top \begin{bmatrix} 0 & -I \\ I & 0 \end{bmatrix} \\ &= \begin{bmatrix} 0 & -I \\ I & 0 \end{bmatrix} \begin{bmatrix} A^\top & C \\ B & -A \end{bmatrix} \begin{bmatrix} 0 & -I \\ I & 0 \end{bmatrix} \\ &= \begin{bmatrix} 0 & -I \\ I & 0 \end{bmatrix} \begin{bmatrix} C & -A^\top \\ -A & -B \end{bmatrix} \\ &= \begin{bmatrix} A & B \\ C & -A^\top \end{bmatrix} \end{align}

Incorporating $J$ into the determinant from earlier allows use to show

\begin{align} \det\left(J(M^\top - \lambda\,I)J\right) &= \det(J)^2\det\left(M^\top - \lambda\,I\right) = \det\left(M^\top - \lambda\,I\right) \\ &= \det\left(J\,M^\top J - \lambda\,J^2\right) \\ &= \det\left(M + \lambda\,I\right) \end{align}

Therefore $\det\left(M - \lambda\,I\right) = \det\left(M + \lambda\,I\right)$. So if $\lambda$ is an eigenvalue $\det\left(M - \lambda\,I\right)$ is zero, but then $\det\left(M + \lambda\,I\right)$ would be zero as well and thus $-\lambda$ has to be an eigenvalue as well.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.