Find eigenvalues of the $(n+1) \times (n+1)$-matrix

$$ \left( \begin {array}{cccccccc} 0&0&0&0&0&0&-n&0\\ 0 &0&0&0&0&-(n-1)&0&1\\ 0&0&0&0&-(n-2)&0&2&0 \\ 0&0&0&\ldots&0&3&0&0\\ 0&0&-3&0&\ldots&0 &0&0\\ 0&-2&0&n-2&0&0&0&0\\ -1&0&n-1&0 &0&0&0&0\\ 0&n&0&0&0&0&0&0\end {array} \right) $$

I have tried for some small $n$ by hand and have got that for $n=3$ the eigenvalues are $1,-1,3,-3$, for $n=4$ eigenvalues are $0,2,-2,4,-4$. So I am conjectured than for arbitrary $n$ the eigenvalues are $n, n-2, n-4, \ldots, -n+2,-n.$

How to prove it?

  • $\begingroup$ mostly you calculate a matrix with columns the eigenvectors, adjusting so they are all integers, and see if there is any pattern that holds up as $n$ increases $\endgroup$ – Will Jagy Dec 4 '18 at 22:40
  • 2
    $\begingroup$ Your matrix, say $M$, belongs to the so-called category of "skew centrosymmetric matrices" $JMJ=-M$ (where $J$ is the antidiagonal matrix). Doesn't know if that helps... $\endgroup$ – Jean Marie Dec 4 '18 at 23:50

Call your matrix $B_{n+1}$. Denote the Kac matrix of the same size by $$ A_{n+1}=\pmatrix{ 0&n\\ 1&0&n-1\\ &2&\ddots&\ddots\\ & &\ddots&\ddots&\ddots\\ & & &\ddots&0 &1\\ & & & & n &0}. $$ It is known that the spectrum of the Kac matrix is given by $$ \sigma(A_{n+1})=\{-n,\,-n+2,\,-n+4,\ldots,\,n-4,\,n-2,\,n\}. $$ We shall prove that $\sigma(A_{n+1})=\sigma(B_{n+1})$. Consequently (as the Kac matrix has distinct eigenvalues) the two matrices are similar to each other.

Here is our proof. First, $B_{n+1}^2$ is similar to $A_{n+1}^2$. In fact, if $D$ denotes the $(n+1)\times(n+1)$ diagonal matrix such that $d_{kk}=(-1)^{\lfloor(k-1)/2\rfloor}$ (i.e. $D=\operatorname{diag}(1,1,-1,-1,1,1,-1,-1,\ldots)$), a straightforward computation will show that $DB_{n+1}^2D=A_{n+1}^2$.

Thus $\sigma(B_{n+1}^2)=\sigma(A_{n+1}^2)$. Yet, as pointed out in the answer by Jean Marie, $B_{n+1}$ is similar to $-B_{n+1}$ via the reversal matrix (the mirror image of the identity matrix from left to right). Therefore, nonzero eigenvalues of $B_{n+1}$ must occur in pairs of the form $\pm\lambda$. Hence we have $\sigma(B_{n+1})=\sigma(A_{n+1})$, because the multiplicity of each nonzero eigenvalue of $A_{n+1}^2$ is exactly $2$.

  • $\begingroup$ Very clever manipulations... $\endgroup$ – Jean Marie Dec 6 '18 at 20:52
  • $\begingroup$ @user1551 Thank you!! $\endgroup$ – Leox Dec 7 '18 at 18:46

This is very far from a complete answer (but too long to fit in a comment).

Let us call $M$ (or $M_n$ if we want to stress the dependency on $n$) the given matrix.

Let us establish the following property : if $\lambda$ is an eigenvalue of $M$, then $-\lambda$ is as well an eigenvalue of $M$. Said otherwise, the spectrum of $M$ is symmetric with respect to $O$.

Let $J$ be the antidiagonal matrix

($J_{ij}=1 \ \iff \ i+j=n+2$ and $J_{ij}=0$ otherwise).

Please note that $J^{-1}=J$.

It is not difficult to establish that

$$JMJ=-M. \tag{1}$$

This is a way to express that matrices like $M$ are "skew-centrosymmetric" (there is some literature on the subject).

Let $I$ be the $n+1$ dimensional unit matrix. Subtracting $\lambda I$ to LHS and RHS of (1), one can write :

$$JMJ-\lambda JIJ=-M-\lambda I $$

Let us left- and -right factorize by $J$ :

$$J(M - \lambda I)J=-(M+\lambda I).$$

Taking determinants of both sides, we get :

$$\det(J)^2\det(M - \lambda I)=(-1)^{n+1}\det(M+\lambda I).$$

Thus, as $\det(J)\neq 0$ (recall that $J^2=I$) :

$$\det(M - \lambda I)=0 \ \ \iff \ \ \det(M-(-\lambda) I)=0,$$

proving the result.


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.