1
$\begingroup$

Consider $n\times 1$ vector $\alpha = (\alpha_{1}, ..., \alpha_{n})$, where $0<\alpha_{i}<1$, and $\sum_{i=1}^{n}\alpha_i = 1$. Construct the $n\times n$ zero-diagonal matrix $A$ with $(i,j)$-th entry

$$A_{ij} = \begin{cases}\frac{\alpha_{i}}{1-\alpha_{j}} & \text{for} \; i\neq j,\\ 0 & \text{otherwise.}\end{cases}$$

I would like to compute (whenever exist) the elements of vector $v = (v_{1}, ..., v_{n})$ as functions of $(\alpha_{1}, ..., \alpha_{n})$, where $Av = v$, $0<v_{i}<1$, and $\sum_{i=1}^{n}v_{i}=1$. In other words, $v$ is the (normalized) eigenvector corresponding to eigenvalue 1. Any idea or suggestions will be appreciated.

$\endgroup$
5
  • 5
    $\begingroup$ Try $\nu_i=\alpha_i(1-\alpha_i)k$ where $k$ is chosen to normalise. $\endgroup$ May 9, 2017 at 8:25
  • $\begingroup$ @ martin: I see. How do you prove it? $\endgroup$ May 9, 2017 at 8:38
  • 2
    $\begingroup$ I wrote it as a system of equations $\frac{\nu_i}{1-\alpha_i}=\sum_j\frac{\alpha_i\nu_j}{1-\alpha_j}$ (here the sum runs over all $j$. Then changed variable $x_i=\nu_i/(1-\alpha_i)$ and observed that $x_i$ has to be proportional to $\alpha_i$. $\endgroup$ May 9, 2017 at 8:41
  • $\begingroup$ @martin: very nice. looks trivial now :-) Thanks. $\endgroup$ May 9, 2017 at 8:52
  • $\begingroup$ @martincripps Want to post it as an answer? $\endgroup$ May 9, 2017 at 13:01

1 Answer 1

4
$\begingroup$

If you write $A\nu=\nu$ as a system of equations it can be written as $$ \frac{\nu_i}{1-\alpha_i}=\sum_{j=1}^{n}\frac{\alpha_i\nu_j}{1-\nu_j} \qquad \forall i=1, ..., n. $$ Then the change of variable $x_i=\frac{\nu_i}{1-\alpha_i}$. This gives the simple system $x_i=\alpha_i\sum_jx_j$, $i=1,\dots,n$, which has a solution $x_i=k\alpha_i$, where $k$ is a normalising constant. Thus $\nu_i=\alpha_i(1-\alpha_i)[\sum_j\alpha_j(1-\alpha_j)]^{-1}$.

$\endgroup$
0

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.