4
$\begingroup$

In Paul.A Clement's (1959) paper:

A Class of Triple-Diagonal Matrices for Test Purposes

SIAM Review, Vol. 1, No. 1 (Jan., 1959), pp. 50-52

He makes the claim that the eigenvalues of :

$ \begin{pmatrix} 0 & y_{1} & 0 & ... & 0 \\\ x_{1} & 0 & y_{2} & & ... \\\ 0 & x_{2} & 0 & ... & 0 \\\ ... & & ... & & y_{n-1} \\\ 0 & ... & 0 & x_{n-1} & 0 \end{pmatrix} $

are $\pm (n), \pm (n-2), ..., (\pm1 \; or \; 0)$ for $x_{k} = k$ and $y_{k} = n-k+1$.

Specifically, and I quote, "then a theorem of Sylvester establishes that the eigenvalues of this An+, are the numbers".

I can't for the life of me figure out what theorem and/or how it follows from them. I am familiar with Sylvester's formula for matrices in terms of their eigenvalues, but to get Frobenius covariants of a matrix A one needs to know the eigenvalues to start with.

Am I overlooking something trivial here?

$\endgroup$
4
  • 1
    $\begingroup$ This website is for questions of research interest. Your question might get a better reception at math.stackexchange.com $\endgroup$ May 29, 2013 at 5:44
  • 2
    $\begingroup$ This is a legit question and I don't see why it will get a better response on M.SE. $\endgroup$ May 29, 2013 at 14:18
  • $\begingroup$ @darij, it certainly got a quicker response at m.se; I won't be drawn on whether it got a better response there. $\endgroup$ May 29, 2013 at 23:49
  • $\begingroup$ (of course, since all I put in my answer there were links to related papers, it can't really be said to be "better"...) $\endgroup$ May 30, 2013 at 19:04

1 Answer 1

3
$\begingroup$

I think the claim in this form is wrong. The eigenvalues are not integral. For example, with $n=4$ the matrix is $$ A=\begin{pmatrix} 0 & 4 & 0 & 0\cr 1 & 0 & 3 & 0\cr 0 & 2 & 0 & 2 \cr 0 & 0 & 3 & 0 \end{pmatrix}. $$

The characteristic polynomial of this matrix is $\chi (t)=t^4-16 t^2 +24$, which has no integral roots. Am I overlooking something ?

Edit: just visited the site http://math.nist.gov/MatrixMarket/deli/Clement/ Here I saw that the upper diagonal must be $3,2,1$, not $4,3,2$. Then everything is OK. So $y_k=n-k$ rather than $y_k=n-k+1$, what you wrote. We have a recursion for $\chi(t)=\det(t\cdot id-A)$, see Eigenvalues of Symmetric Tridiagonal Matrices.

Edit: A proof for the correct claim can be found in the paper of Taussky and Todd, "Another look at a matrix of Mark Kac", Linear Algebra Appl. 150 (1991), 341-360. More details are also at https://math.stackexchange.com/questions/405670/regarding-a-paper-by-paul-a-clement-on-tridiagonal-matrices, which was remarked by Gerry (thank you ). Also, Darij's remark is correct, that the proof does not follow easily, so I have edited this here.

$\endgroup$
4
  • $\begingroup$ Why easily? Where is the symmetry? $\endgroup$ May 29, 2013 at 20:46
  • $\begingroup$ Yes, OP posted to m.se, I posted a $3\times3$ counterexample, then the correction to the problem was noted, and a link was given to a previous m.se question about the same matrix. $\endgroup$ May 29, 2013 at 23:47
  • $\begingroup$ @Darij: you are right, the proof is not evident. My answer was (is) a comment about the mistake of the matrix, not more. @Gerry: I have found your link and the answers, thank you. The correction to the problem got lost, as we see. $\endgroup$ May 30, 2013 at 11:58
  • 1
    $\begingroup$ As you have linked to the paper by Taussky and Todd, you might want to mention www-math.mit.edu/~edelman/homepage/papers/kac.pdf as well... $\endgroup$ May 30, 2013 at 19:03

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.