1
$\begingroup$

How can I construct matrices with positive integer values and distinct integer eigenvalues (not necessarily positive, but 0 should not be an eigenvalue). The standard-method to construct matrices with given eigenvalues is a transformation $A\rightarrow B^{-1}AB$, but the problem is that such transformations do not tend to produce positive matrices. A few examples are (the eigenvalues are shown in brackets)

$[4, 2; 3, 5] \ \ \ \ [2, 7]~$

$[2, 5, 2; 5, 2, 6; 6, 6, 5] \ \ \ [-3, -1, 13]~$

$[1, 5, 4, 3; 4, 6, 1, 2; 5, 1, 4, 3; 2, 5, 4, 2] \ \ \ [-3, -1, 4, 13]~$

$[1, 1, 5, 3, 5; 4, 6, 3, 4, 5; 6, 6, 5, 5, 4; 6, 3, 5, 5, 4; 5, 6, 4, 5, 4] $

$[-3, -1, 1, 2, 22]~$

I do not know a 6x6-matrix or a bigger one with the desired property.

$\endgroup$
3
$\begingroup$

For a $6 \times 6$ example, try $$ \left[ \begin {array}{cccccc} 6&1&1&1&1&1\\ 1&6&1&1 &1&1\\ 1&2&5&1&1&1\\ 1&2&2&3&1&2 \\ 1&2&2&1&3&2\\ 1&2&2&1&2&3 \end {array} \right] $$ with eigenvalues $1,2,3,4,5,11$.

EDIT: More generally, start with an $n \times n$ diagonal matrix $D$ with positive integer diagonal entries $d_1 > d_2 > \ldots > d_n$. Let $B$ be the $n \times n$ matrix with first row $[d_1, d_2, \ldots, d_n$ and all else $0$. Thus $B + D$ is upper triangular with eigenvalues $2 d_1, d_2, \ldots d_n$ which are distinct positive integers. Take $A = L (B + D) L^{-1}$ where $L$ is the lower triangular matrix with all entries $1$ on and below the main diagonal. Of course, this has the same eigenvalues as $B+D$. Note that $L^{-1}$ has $1$ on the main diagonal, $-1$ on the diagonal below that, and everything else $0$. Then $$A_{ij} = \cases{d_j - d_{j+1} & if $ i < j < n$\cr 2 d_j - d_{j+1} & if $ i = j < n$\cr 2 d_j - 2 d_{j+1} & if $i > j$\cr d_n & if $i < j = n$\cr 2 d_n & if $i=j=n$\cr }$$ which are all positive integers.

EDIT: Still more generally, you don't need to use the same positive integers in $B$ and $D$...

$\endgroup$
  • $\begingroup$ Very nice, how did you get it ? $\endgroup$ – Peter Jul 6 '14 at 18:50
  • $\begingroup$ It even has POSTIVE eigenvalues. I see directly, that 1,2 and 5 are eigenvalues. I checked with PARI, that the others also are. $\endgroup$ – Peter Jul 6 '14 at 18:57
  • $\begingroup$ Wow! Very instructive. Thank you very much! $\endgroup$ – Peter Jul 6 '14 at 19:18
  • $\begingroup$ This method allows you to take an arbitrary set of $n$ distinct positive integers, and make them the eigenvalues of an $n\times n$ matrix with positive entries. Nice! $\endgroup$ – vadim123 Jul 6 '14 at 19:33
  • $\begingroup$ Not quite, $d_1$ has to be doubled. $\endgroup$ – Peter Jul 6 '14 at 19:34

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.