10
$\begingroup$

I'm studying the following tri-diagonal matrix

$$ X = \begin{pmatrix} 0 & x_0 & 0 & 0 &\cdots & 0 & 0 & 0 \\\ x_0 & 0 & x_1 & 0 &\cdots & 0 & 0 & 0 \\\ 0 & x_1 & 0 & x_2 &\cdots & 0 & 0 & 0 \\\ \vdots & \vdots & \ddots & \ddots & \ddots & \vdots & \vdots& \vdots \\\ \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & \vdots& \vdots \\\ \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots& \vdots \\\ 0 & 0 & 0 & 0 &\cdots & x_{N-1} & 0 & x_N \\\ 0 & 0 & 0 & 0 & \cdots & 0 & x_N & 0 \end{pmatrix} $$

where the entries are given by the sequence $x_n = \frac{\sqrt{(N+1-n)(n+1)}}{N+1}$

Numerically I find that the eigenvalues are given by $\lambda_k = 1-\frac{2k}{N+1}$ for $k\in[0,N+1]$.

Also I find numerically that the orthogonal transformation $V$ diagonalizing the above matrix $D = V^T X V$ (such that the columns of $V$ are the eigenvectors of $X$) can be chosen to be symmetric: $V = V^T$.

I have been trying to find/come up with a proof for the above two statements, however, so far unsuccessful. I do have an exact expression for the eigenvector $\vec{v}_0 = c_i\vec{e}_i$ corresponding to $\lambda_0=1$, being $$ c_i = \frac{1}{\sqrt{2^{N+1}}}\sqrt{\begin{pmatrix} N+1 \\ i \end{pmatrix} } $$

Also $X = V^TDV$ implies a recurrence relation on the eigenvectors $\vec{v}_i$ of $X$, being $\lambda\vec{v}_k = x_{k-1}\vec{v}_{k-1} +x_k\vec{v}_{k+1}$ (this is equivalent the statement $V=V^T$). Unfortunately this has not helped me so far.

Any help on proving either of the above statements would be greatly appreciated!

Thanks.

$\endgroup$
  • $\begingroup$ So for $N=0$ you think the matrix has eigenvalues $\pm 1?$ $\endgroup$ – Aaron Meyerowitz Mar 23 '16 at 20:01
  • 1
    $\begingroup$ For $N=0$ the matrix is $\begin{pmatrix} 0 & 1\\ 1&0\end{pmatrix}$ so that checks out. $\endgroup$ – Nick Cook Mar 23 '16 at 22:43
  • 1
    $\begingroup$ I'm surprised at the close votes. This looks like a perfectly reasonable question to me, probably well above MO average. $\endgroup$ – Christian Remling Mar 24 '16 at 0:11
  • $\begingroup$ unfortunately, I misread the question so clicked 'close', but it seems like an interesting question... $\endgroup$ – Suvrit Mar 24 '16 at 1:36
4
$\begingroup$

The Matrix $X$ I was looking at was nothing more than the angular momentum operator $L_x$ in the spin $l=(N+1)/2$ representation, see angular momentum on wikipedia.

It can be written as $L_x = 1/2(L_+ + L_-)$ where $L_+$ and $L_-$ obey $[L_z,L_\pm] = L_\pm$ and $L_z$ is the diagonal matrix with integer spaced entries ranging from $-l$ tot $l$. From the knowledge of spin representations (or $SL(2,\mathbb{C}$) if you which) one can easily find the spectrum of $L_x$.

Also, the matrix diagonalizing $X$ is just $V = e^{\pi i/2 L_y}$ which in itself is not symmetric but from it a matrix $V' = VD$ can be constructed which also diagonalizes $X$ and which is symmetric. Here $L_y = i/2(L_+ - L_-)$.

I would like to thank all who have put thought in my question!

$\endgroup$

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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