2
$\begingroup$

I have been trying to solve the following difference equation for some time now : $$u^3(n+1) = a - b\cdot u^2(n) + u^3(n), \qquad a \ne 0 \ne b$$ I have tried various substitutions, simplifications but nothing did.

My last attempt was to transform it into a continuous differential equation by letting $u(n) = u(t_n)$, with $t_n = h + t_{n-1}$ and $t_0 = 0$ ie $t_n = h\cdot n$. From there I transformed the above equation in : $$\frac{u^3(t_{n+1}) - u^3(t_n)}{h} = \frac{a - b\cdot u^2(t_n)}{h}$$ $$\frac{u(t_{n+1}) - u(t_n)}{h} \cdot (u^2(t_{n+1}) + u(t_{n+1})\cdot u(t_n) + u^2(t_n)) = \frac{a - b\cdot u^2(t_n)}{h}$$ Taking the limit as $h\longrightarrow 0$, I get $$3u^2\cdot\frac{du}{dt_n} = \frac{a - b\cdot u^2}{h}$$

From here I can solve $\int\frac{3h\cdot u^2}{a-b\cdot u^2} du = \int dt_n$,

integrating gives me $$\frac {t_n} h = \frac{3\sqrt a \cdot tanh^{-1}(u\cdot \sqrt{\frac ba})}{b^{3/2}} - \frac{3u}{b} + C$$ but I'm interested in the inverse of this function, as in $u(t_n) = ...$

Does anybody know how to either solve anatically the initial difference equation or find an approximation?

Also is my continuous approach any good and is it possible to invert it? I tried using a polynomial expansion but it wouldn't converge quickly enough.

$\endgroup$
  • 1
    $\begingroup$ I guess * means multiply? This is a discrete dynamical system - probably no explicit solution, and the tag "dynamical-systems" should be added. $\endgroup$ – user25199 Jul 31 '13 at 12:32
  • $\begingroup$ The word "solve" has different meanings for different people. For instance, what is wrong with selecting your favorite initial condition and obtaining next, say, million steps on a computer? If you want certain limiting or qualitative information about the solutions, you should be more specific. As Carl already mentioned, a simple formula for the general solution is unlikely to exist. $\endgroup$ – Igor Khavkine Jul 31 '13 at 19:04
  • $\begingroup$ @Igor A million steps on a computer can be quite misleading for 1D maps and ordinary precision (say, 16 digits). There is a significant probability of finding a "numerical" periodic orbit, even if all periodic orbits in the real system are unstable; the classic example is the Ulam map $x\to4x(1-x)$ for which most initial conditions reach $x=0$ at remain there after a few million iterations. $\endgroup$ – user25199 Aug 1 '13 at 10:18
  • $\begingroup$ perhaps the Carleman linearization method can be usefull but the linearized form may become an infinte system of equations. $\endgroup$ – user36539 Sep 19 '13 at 16:14
1
$\begingroup$

Some ideas: You can find approximate solutions near the fixed points, $\pm\sqrt{a/b}$ (if real) and $\pm\infty$. The latter are probably good places for your continuum approximation.

Then, there is strong contraction when the derivative of the RHS of the first equation is zero, ie $0$ and $2b/3$. Depending on the parameters this may lead to stable periodic orbits, and/or may determine behaviour of chaotic orbits - read up on kneading theory.

But I doubt if there is a closed form solution, for either the original equation or for the desired inverse.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Wouldn't you need to cubic root the RHS and then take the derivative to analyse whether it's stable or not? $\endgroup$ – user88595 Aug 5 '13 at 9:26
  • $\begingroup$ For the fixed points (first paragraph of my answer), you have the right approach. For the periodic orbits (second paragraph), if a critical point is periodic it is superstable (has zero derivative), and there will be a range of parameters near it where there is a stable periodic orbit. But the exact location is a solution of a high degree polynomial and so more difficult to analyse exactly. $\endgroup$ – user25199 Aug 19 '13 at 8:16

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.