Let $A$ be a matrix of order $ n \times n$ and let $ \lim_{ k \rightarrow \infty} A^k = L > 0.$ I want to show that there exists some power $m$ such that $A^m>0$ and $ A^{m+i} > 0$ for any $ i \geq m.$ By a matrix $L>0$ and $A>0$ I mean all entries of the matrices are positive. Any hint or a proof please?

Thank you in advance.


Since $\lim\limits_{k\to\infty} A^k=L$, then this holds for all entries of matrix $A$. In other words $$ \lim\limits_{k\to\infty} A_{ij}^k=L_{ij}>0\tag{1} $$ for all $i$ and $j$. Now take arbitrary $i,j$, then from $(1)$ it follows that there exist $m_{ij}\in\mathbb{N}$ such that $k\geq m_{ij}$ holds $A_{ij}^k>0$. Now consider $m=\max_{i,j} m_{ij}$, then for all $k>m$ and all $i,j$ we have $A_{ij}^k>0$. Which means that there exist $m\in\mathbb{N}$ such that for all $k\geq m$ holds $A^k>0$

  • $\begingroup$ If we consider a matrix $ A = \begin{bmatrix} 0&0.1&0.5\\0.2&0&0\\0&0.3&0 \end{bmatrix}$ we see that $A^5>0$ but some $A_{ij}^5 =0$ for some $i,j.$ In fact $A_{ij}^k = 0$ for any power $k$ for some entries $A_{ij}$ of $A.$ So how did you get Equation(1)? $\endgroup$ – Zizo Nov 13 '12 at 10:58
  • $\begingroup$ $$A=\left( \begin{array}{ccc} 0.0012 & 0.00454 & 0.0002 \\ 0.00008 & 0.0012 & 0.003 \\ 0.0018 & 0.00012 & 0.0006 \end{array} \right)$$ so all the entries are positive. So something wrong with your computations $\endgroup$ – Norbert Nov 13 '12 at 11:02
  • $\begingroup$ Thanks! I now understood what you mean, there was miss reading because I read it $[A_{ij}]^k$ not $[A^k]_{ij}.$ I just cannot see when the entries $A_{ij}^m>0$, then $A^{k}>0$ for any $ k > m.$ $\endgroup$ – Zizo Nov 13 '12 at 11:24
  • $\begingroup$ Okay okay! I think I can see it! $\endgroup$ – Zizo Nov 13 '12 at 11:27
  • $\begingroup$ How $[A^k]_{ij}>0$ for any $k \geq m$ lead to $A^k>0$ while $A^k$ might does not have the entries $[A^k]_{ij}.$ I am just thinking that $A^k = A^{k-m}A^m$ so $A^{k-m}$ might have entries which are zero, in other words is it always true when you multiply a matrix $B \geq 0$ by $C>0$ then $BC>0.$ Is not always true I think. $\endgroup$ – Zizo Nov 13 '12 at 11:35

For $1 \leq i,j \leq n$, consider $\pi_{ij}$, the map which project a matrix on its $(i,j)$ component. The set of positive matrices in the sense you define is in fact $$\bigcap_{1\leq i,j\leq n}\pi_{ij}^{-1}(\mathbb R_0^+).$$ Since $\pi_{ij}$ is continuous, this is open (finite intersection of open subsets), hence the result.

  • $\begingroup$ Can you elaborate on this? Can you show that such matrices are an open set. I know the equivalence of norms in matrices space. How would this help me? $\endgroup$ – Zizo Nov 13 '12 at 11:45
  • $\begingroup$ I edited my answer to show that the set is indeed open.Equivalence of norms is of no particular importance here, but it is a good thing to know when speaking of topology in finite dimensional vector spaces. $\endgroup$ – YoungFrog Nov 15 '12 at 13:10

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.