MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Hi all,

do you know how to compute (as a function of n) the largest eigenvalue of this matrix (or at least to bound it)?

$$ \left(\begin{array}{cccccc} 0 & 1 & & & & \cr 1 & 0 & \sqrt 2 & & & \cr & \sqrt 2 & 0 & & & \cr & & & \ddots & & & \cr & & & & 0 & \sqrt n & \cr & & & & \sqrt n & 0 & \end{array}\right) $$


share|cite|improve this question
You can bound it by Gershgorin. out of interest, does this come from some kind of random matrix problem? – Yemon Choi Feb 14 '12 at 16:40
I ask because this matrix is the expected value of the natural tridagonalization of the unnormalized GOE, although that may not be relevant here – Yemon Choi Feb 14 '12 at 16:55
you can bound it above $2\sqrt{n}$. – Kate Juschenko Feb 14 '12 at 17:50
@Yemon: It looks to me like the sum of a creation and an annihilation operator on (a finite-dimensional piece of) a Fock space in quantum field theory. Would that have any connection with the GOE that you mentioned? – Andreas Blass Feb 14 '12 at 19:03
The dominant eigenvalue is asymptotic to $2 \sqrt n$. This can be seen by comparison with multiples of the $m \times m$ matrix, call it $L_m$, whose $(i,j)$ entry is $1$ if $|i-j|=1$ and $0$ else. The eigenvalues of $L_m$ are known explicitly and the largest one approaches $2$ from below as $m \rightarrow \infty$. Kate Juschenko may have used this to obtain the upper bound $2 \sqrt n$ (the matrix is smaller than $\sqrt n M_{n+1}$). – Noam D. Elkies Feb 14 '12 at 19:52
up vote 11 down vote accepted

If you denote $A_n$ your tri-diagonal matrix of order $n$, and $H_n(x):= \det(x+A_n)$, the sequence $H_n$ satisfies the two-term linear recurrence $H_{n+1}=xH_n - nH_{n-1}$ with initial conditions $H_0=1$ and $H_1=x$. Thus, they are the Hermite polynomials (here in the "probabilist's version"), and their zeros are the eigenvalues of $-A_n$ (on which you can find everything in the literature).

share|cite|improve this answer
I'd like to add that the connection with random matrices that Yemon Choi suggests above may be the following : The mean characteristic polynomial of a $n\times n$ GUE random matrix X, namely $\mathbb{E}_{GUE}\det(xI_n−X)$, is actually the Hermite polynomial (this formula goes back to Heine), for which the matrix of recurrence coefficients is given by An, as you pointed out. Even so I don't really see GOE there, even it is plausible. The matrix An is also interpreted as a Toeplitz operator associated with the sum of a creation and annihilation operator as pointed out by Andreas Blass. – Adrien Hardy Feb 14 '12 at 23:34
The connection with such creation/ annihilation operator has been used by Voiculescu once developing free probability theory. I think this is explained in the book of Hiai and Petz, "The Semicircle Law, Free Random Variables and Entropy". I hope this helps ... – Adrien Hardy Feb 14 '12 at 23:34
@Adrien: see my comments to the main question for some attempts to correct and clarify my initial claims – Yemon Choi Feb 15 '12 at 0:11

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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