I have a block matrix

$$M=\begin{bmatrix} I_0& I_1& \cdots& I_1\\ I_2& I_0& \ddots& \vdots\\ \vdots& \ddots& \ddots& I_1\\ I_2& \cdots& I_2& I_0\\ \end{bmatrix}_{n \times n}$$


$$I_0=\begin{bmatrix} 0& 1\\ 1& 0\\ \end{bmatrix}, \qquad I_1=\begin{bmatrix} 0& 1\\ -1& 0\\ \end{bmatrix},\qquad I_2=\begin{bmatrix} 0& -1\\ 1& 0\\ \end{bmatrix}.$$

I want to find all its eigenvalues $\{\lambda_1,\lambda_2,\ldots,\lambda_{2n}\}$, where $\lambda_1 < \lambda_2 < \cdots < \lambda_{2n}$.

Due to the chiral symmetry, we can find $\lambda_i=-\lambda_{2n+1-i}$ for all $i$.

  • 1
    $\begingroup$ I don't know about the chiral symmetry. But $I_0^2=I$, $I_1^2=-I$, and $I_0I_1+I_1I_0=0$ (where $I$ is the $2\times2$ identity matrix) imply $M^2=-(n-2)E$, where $E_{i,j}=I$ for all $i,j$. Hence $M^2/(-n(n-2))$ is an orthogonal projection of rank $2$. This means that the nonzero eigenvalues of $M^2$ is only $-n(n-2)$ with multiplicity $2$. The nonzero eigenvalues of $M$ are simple and $\pm i\sqrt{n(n-2)}$ (because the trace is zero). $\endgroup$ Mar 2, 2023 at 9:10
  • $\begingroup$ @NarutakaOZAWA I am so sorry. I ignore an important property, i.e., M=M^T. $\endgroup$
    – Young Q
    Mar 2, 2023 at 11:35
  • $\begingroup$ It can be shown that finding the eigenvalues of $M$ is equivalent to finding those of $A$ or $A^2$ where $A$ is the matrix with a zero diagonal and entries $a_{i,j}=1$ for $j>i$; $a_{i,j}=-1$ for $i>j$. If you try with $n$ you see the eigenvalues (for $A$ and $M$) gets complicated with radicals in terms of (large) $n$. $\endgroup$
    – Toni Mhax
    Mar 2, 2023 at 15:19

1 Answer 1


For the signed circulant matrix $$U:=\left[\begin{matrix} & 1 & & & \\ & & \ddots & & \\ & & & 1 &\\ -1 & & & & \end{matrix}\right] \mbox{ in } M_n(\mathbb{C}),$$ one has $$M= 1 \otimes I_0 + (U+U^2+ \cdots + U^{n-1}) \otimes I_1 \mbox{ in } M_n(\mathbb{C})\otimes M_2(\mathbb{C}).$$ For $\omega:=\exp\frac{i\pi}{n}$, the unitary matrix $U$ has eigenvalues $\{ \omega^k : k=1,3,5,\ldots,2n-1\}$ and eigenvectors $v_k:=[\begin{smallmatrix} 1 & \omega^k & \omega^{2k} & \cdots &\omega^{(n-1)k}\end{smallmatrix}]^{\mathrm{T}}/\sqrt{n}$ in $\ell_2^n$. Accordingly, the matrix $M$ is decomposed into the direct sum of $$I_0+ (\omega^k+\omega^{2k}+ \cdots + \omega^{(n-1)k})I_1 =\left[\begin{matrix} 0 & \lambda_k\\ \overline{\lambda_k} & 0 \end{matrix}\right] \mbox{ acting on } \mathbb{C}v_k \otimes \ell_2^2,$$ where $$\lambda_k=\frac{2}{1-\omega^k}.$$ The eigenvalues of $M$ are $\pm|\lambda_k|$, $k=1,3,\ldots,2(n-1)$ with eigenvectors $ v_k \otimes [\begin{smallmatrix} 1 & \pm \mathrm{sgn}(\overline{\lambda_k}) \end{smallmatrix}]^{\mathrm{T}}/\sqrt{2}$ in $\ell_2^n \otimes \ell_2^n$. Note that since $|\lambda_k|=|\lambda_{2n-k}|$, all eigenvalues except for $1$ (corresponding to the case when $n$ is odd and $\lambda_n=-1$) have multiplicity $2$.

  • $\begingroup$ I have two questions. 1. Can we use the method to solve the eigenvalues of any Toeplitz matrices? 2. Can we solve its eigenstates? $\endgroup$
    – Young Q
    Mar 3, 2023 at 9:03
  • 2
    $\begingroup$ I revised my answer to include eigenvectors. The answer to your question 1 is negative. There is no general methods of calculating eigenvalues of a general (hermitian) Toeplitz matrices (and there is a lot of research on this topics). $\endgroup$ Mar 3, 2023 at 14:19
  • $\begingroup$ I understand. If the diagonal terms (i.e., I_0) are not identical, is the method is ineffective? Here, I_0, I_1, and I_2 can be regarded as the numbers. $\endgroup$
    – Young Q
    Mar 5, 2023 at 16:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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