7
$\begingroup$

Find eigenvectors of the $(n+1) \times (n+1)$-matrix: $$\left(\begin {array}{cccccccc} 0&0&0&0&0&0&-1&0\\ 0&0&0&0&0&-2&0&n\\ 0&0&0&0&-3&0&n-1&0 \\ 0&0&0&\ldots&0&n-2&0&0\\ 0&0&-(n-2)&0&\ldots&0 &0&0\\ 0&-(n-1)&0&3&0&0&0&0\\ -n&0&2&0 &0&0&0&0\\ 0&1&0&0&0&0&0&0\end {array} \right)$$

The eigenvalues of the matrix are $ n, n-2, n-4, \ldots, -(n-2), -n$ see the question

Let $e_{\lambda}$ denotes the eigenvector corresponding to the eigenvalues $\lambda$. So far I have found

$$ e_0=[a_0, a_1, \ldots, a_n], a_i=\begin{cases} 0, \text{ $i$ odd}\\ \displaystyle\binom{\frac{n}{2}}{ \frac{i}{2}}, \text{$i$ even} \end{cases} $$ For example for $n=8$ we have $e_0=[1, 0, 4, 0, 6, 0, 4, 0, 1]$.

Also $$ e_1=[a_0, a_2, \ldots, a_n], a_i=\begin{cases} 0, \text{ $i$ even}\\ \displaystyle\binom{\frac{n-1}{2}}{ \frac{i-1}{2}}, \text{$i$ odd} \end{cases}. $$ for example for $n=7$ we have $e_1=[0, 1, 0, 3, 0, 3, 0, 1]$, and

$$ e_2=[a_0, a_2, \ldots, a_n], a_i=\begin{cases} \displaystyle \binom{\frac{n}{2}-1}{\frac{i}{2}-1}-\binom{\frac{n}{2}-1}{\frac{i}{2}}, i\text{ even} \\ \\ \displaystyle 2(-1)^{i-1}\binom{\frac{n}{2}-1}{\frac{i-1}{2}} , i\text{ odd} \end{cases}. $$

for example for $n=8$ we have $e_2=[-1, 2, -2, 6, 0, 6, 2, 2, 1]$.

Question: What is an eigenvector $e_{\lambda}$ (with integer coordinates) for arbitrary $\lambda \in \{ n, n-2, n-4, \ldots, -(n-2), -n\}$?

I hope the problem was already solved in 19th century.

$\endgroup$
  • 1
    $\begingroup$ Little remark : I would say "an" eigenvector (with integer coordinates) instead of "the" eigenvector, because it is defined up to a multiplicative constant. $\endgroup$ – Jean Marie Dec 8 '18 at 20:58
  • $\begingroup$ It looks like a too specific matrix theory problem, so if you will not obtain an answer at MSE then I would recommend you to ask it at MathOverflow. $\endgroup$ – Alex Ravsky Dec 9 '18 at 18:49
3
$\begingroup$

J.M. has commented in a Kac matrix question that the eigenvector matrix for the Kac matrix are given by the (zero-indexed) $(n+1)\times(n+1)$ "Krawtchouk matrix" $K$ defined by $$ K_{ij}=\sum_{k=\max(0,\,i+j-n)}^{\min(i,j)} (-1)^k\binom{j}{k}\binom{n-j}{i-k},\quad (i,j\in\{0,1,\ldots,n\}). $$ Since your matrix is similar to the Kac matrix, we expect that the coefficients of its eigenvectors are closely related to $K_{ij}$. If you try a few $n$s, you will see that some patterns do emerge. Apparently, an eigenvector matrix $V$ is given by $$ V_{ij}= \begin{cases} (-1)^\left\lfloor(i+j+\color{red}{2})/2\right\rfloor\left(\frac{1\color{red}{-}(-1)^{i+j}}2\right)K_{ij} &\text{ when }n\equiv1 (\!\!\!\mod 4),\\ (-1)^\left\lfloor(i+j+\color{red}{1})/2\right\rfloor K_{ij} &\text{ when }n\equiv2 (\!\!\!\mod 4),\\ (-1)^\left\lfloor(i+j+\color{red}{1})/2\right\rfloor\left(\frac{1\color{red}{+}(-1)^{i+j}}2\right)K_{ij} &\text{ when }n\equiv3 (\!\!\!\mod 4),\\ (-1)^\left\lfloor(i+j+\color{red}{2})/2\right\rfloor K_{ij} &\text{ when }n\equiv0 (\!\!\!\mod 4). \end{cases} $$ To prove the correctness of this formula, one may verify that (if we call your matrix $B$) $BV=V\operatorname{diag}(n,n-2,\ldots,2-n,-n)$. This amounts to proving some identities that involve binomial coefficients. I haven't mathematically proved it, but I have verified that the formula is indeed correct for $1\le n\le 20$ using the Matlab program below:

function [B,V,D,diff]=kac(n)
% we should have B*V=V*D

B=fliplr(diag(n:-1:1,-1)-diag(1:n,1));
D=diag(n:-2:-n);

K=zeros(n+1,n+1);
for i=0:n
  for j=0:n
    for k=max(0,i+j-n):min(i,j)
      K(i+1,j+1)=K(i+1,j+1)+(-1)^k*nchoosek(j,k)*nchoosek(n-j,i-k);
    end
  end
end

r=rem(n,4);
if r==1
  S=(-1).^floor((2+repmat((0:n)',1,n+1)+repmat(0:n,n+1,1))./2);
  Z=(1-(-1).^(repmat((0:n)',1,n+1)+repmat(0:n,n+1,1)))./2;
  V=K.*S.*Z;
elseif r==2
  S=(-1).^floor((1+repmat((0:n)',1,n+1)+repmat(0:n,n+1,1))./2);
  V=K.*S;
elseif r==3
  S=(-1).^floor((1+repmat((0:n)',1,n+1)+repmat(0:n,n+1,1))./2);
  Z=(1+(-1).^(repmat((0:n)',1,n+1)+repmat(0:n,n+1,1)))./2;
  V=K.*S.*Z;
else
  S=(-1).^floor((2+repmat((0:n)',1,n+1)+repmat(0:n,n+1,1))./2);
  V=K.*S;
end

diff=max(max(abs(B*V-V*D)))  % should be zero

return

Matlab doesn't seem to use big integers to calculate binomial coefficients. So, the above program may return a false negative result if $n$ is large.

$\endgroup$
  • $\begingroup$ Yes, you was right and I prove it now. Thank you! $\endgroup$ – Leox Dec 25 '18 at 11:17

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.