13
$\begingroup$

$s_{n+1}=s_n^2-s_{n-1}^2$, $s_0=\sqrt{x}$, $s_1=x$
This sequence seems simple, but is pretty confusing. If you try it with integers, you might think that it always diverges to infinity, but if you try something like $1.8,$ it converges to $0.$ Also, this system is chaotic. As an example, $1.670718$ diverges but $1.670719$ converges.

What I'm wondering is:
What is the lowest and highest value that converges?
Does it ever converge to a value other than $0$?

This image may help (The sequence plotted how the Mandelbrot Set would be) The function for complex numbers

$\endgroup$
12
  • 5
    $\begingroup$ If the system is chaotic then how can you tell whether it actually converges or diverges? It seems likely that numerical errors accumulate, and might shift it from diverging to converging or the other way around $\endgroup$ Commented May 15 at 3:14
  • 3
    $\begingroup$ Using a thousand digits of precision it seems like both 1.670718 and 1.670719 converge, although 1.7 still diverges, so it's not monotonic $\endgroup$ Commented May 15 at 3:23
  • 11
    $\begingroup$ Does it ever converge to a value other than 0? - no, if the limit equals $a$, we get $a=a^2-a^2=0$ $\endgroup$ Commented May 15 at 4:34
  • 5
    $\begingroup$ So do you want to study the recursion in $\mathbb R$ or in $\mathbb C$? $\endgroup$ Commented May 15 at 6:31
  • 8
    $\begingroup$ I think that the requirement $s_0 = \sqrt{x}$ is a red herring. You really should be looking at the dynamics of the map $F: (x,y) \to (y, y^2 - x^2)$ on $\mathbb R^2$. It is clear that $(0,0)$ is the only fixed point, and it is an attractor. There is an unstable $3$-cycle $(1,0)$, $(0,-1)$, $(-1,1)$. $\endgroup$ Commented May 15 at 19:39

2 Answers 2

9
$\begingroup$

Here is my approximation to the basin of attraction of $(0,0)$ for the map $F: (x,y) \to (y, y^2 - x^2)$ (shown in black).

enter image description here

$\endgroup$
3
$\begingroup$

Let $F$ be the dynamical system $(x,y)\mapsto(y,y^2-x^2)$. The orbit of any point $(x,y)$ with $|x|,|y|<1$ converges to the origin. In particular, the sequence $\{s_n\}_{n=0}^\infty$ converges to $0$ if $s_0<1$.

Proof) Fix $x_0,y_0\in (-1,1)$. Denote the elements of its orbit under $F$ by $(x_n,y_n):=F^{\circ n}(x_0,y_0)$. We shall show that $\lim_{n\to\infty}x_n=\lim_{n\to\infty}y_n=0$.

Pick an arbitrary number $a$ from $\big(\max(|x_0|,|y_0|),1\big)$. Thus $|x_0|,|y_0|<a$. Now let $\epsilon$ be an element of $\left(\max\left(\sqrt{a},\frac{|y_0|}{a}\right),1\right)$. Hence $|y_0|<\epsilon a$, and $a<\epsilon^2$. I use induction to show that $|x_n|\leq\epsilon^n a$ and $|y_n|\leq\epsilon^{n+1}a$ for any $n\geq 0$. This will clearly yield the desired limits. The base of the induction, the case of $n=0$, holds because of the way we chose $\epsilon$ and $a$. Suppose the induction hypothesis holds: $|x_n|\leq\epsilon^n a, |y_n|\leq\epsilon^{n+1}a$ for some $n\geq 0$. Next, we have $(x_{n+1},y_{n+1})=(y_n,y_n^2-x_n^2)$. Thus: $$ \begin{split} &|x_{n+1}|=|y_n|\leq\epsilon^{n+1}a,\\ &y_{n+1}\leq y_n^2\leq\epsilon^{2n+2}a^2\stackrel{\text{due to }\epsilon,a<1}{<}\epsilon^{n+2}a,\\ &y_{n+1}\geq -x_n^2\geq -\epsilon^{2n}a^2\stackrel{\text{ due to }\epsilon<1\text{ and }a<\epsilon^2}{>} -\epsilon^{n+2}a. \end{split} $$ We conclude that $|x_{n+1}|\leq\epsilon^{n+1}a$ and $|y_{n+1}|\leq\epsilon^{n+2}a$.

Finally, notice that terms of the sequence $\{s_n=s_n(x)\}_{n=0}^\infty$ satisfy $F(s_{n-1},s_n)=(s_n,s_{n+1})$. Consequently, $(s_{n-1},s_n)\to(0,0)$ if $(s_0=\sqrt{x},s_1=x)$ lies in $(-1,1)^2$. $\quad\blacksquare$

$\endgroup$
2
  • 5
    $\begingroup$ Thank you for writing up my comment so that I don’t have to. $\endgroup$ Commented May 16 at 7:28
  • $\begingroup$ @EmilJeřábek I didn't mean to write up your comment. I didn't notice it among all the comments to the original question. The picture of the basin of attraction motivated me to show that $(-1,1)^2$ is contained in the basin. I would happily delete my answer if you are planning to write one yourself. $\endgroup$
    – KhashF
    Commented May 16 at 13:10

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .