I need to study the behavior of this discrete dynamical system:

$$\left\{ \begin{array}{c} x_{n+1}= y_n \\ y_{n+1}= \frac{1}{a}x_n-\frac{1}{a}y_n \\ \end{array} \right.$$

in function of the parameter $a>0$.

I studied before that the stability of the origin is given by the eigenvalues: $\lambda_1 , \lambda_2$

If $\lambda_1 , \lambda_2 < 1$, the origin is stable, hence, attractor.

If $\lambda_1 , \lambda_2 > 1$, the origin is unstable.

If $\lambda_1 > 1 > \lambda_2$ or $\lambda_2 > 1 > \lambda_1$, the origin is a saddle point.

First I found the matrix $A$ associated to this system

$A := \begin{bmatrix} 0 & 1 \\ \frac{1}{a} & \frac{-1}{a} \end{bmatrix}$

But I have problems when I'm studying the eigenvalues for stability. After analyzing the three cases, I determinated that the origin is a saddle point if $a \in (0,2)$ or the origin is stable if $a \in (2,+\infty)$. The problem is that if I graph this here, I always get a saddle point (at least in all the values of $a>0$ that i tested.

Any hints?

  • $\begingroup$ Is $a > 0$ only? $\endgroup$
    – user7440
    Aug 20, 2019 at 5:35
  • $\begingroup$ @user7440 yes, edited. Sorry about that. $\endgroup$
    – Trobeli
    Aug 20, 2019 at 5:54
  • $\begingroup$ Note: This question and this question $\endgroup$
    – Axion004
    Aug 20, 2019 at 17:38

3 Answers 3


This system can be represented as

$$ U_n = M U_{n-1} $$

with $U_n = (n_n, y_n)^{\dagger}$ and $M = \left(\begin{array}{cc}0 & 1\\ \frac 1a & -\frac 1a\end{array}\right)$

but $M = T^{-1}\Lambda T$ then

$$ U_n = T^{-1}\Lambda T U_{n-1}\Rightarrow TU_n = \Lambda T U_{n-1} $$

calling now $V_n = TU_n$ we have

$$ V_{n+1} = \Lambda V_n $$

Here $\Lambda = \left(\begin{array}{cc}-\frac 12\left(\frac{1+\sqrt{1+4a}}{2a}\right) & 0\\ 0 & \frac 12\left(\frac{-1+\sqrt{1+4a}}{2a}\right)\end{array}\right)$ the diagonal eigenvalues matrix. Those eigenvalues have opposite signs characterizing a saddle point.

$$ V_n = \Lambda^n V_0 $$

Attached a plot showing in blue the two eigenvalues as well as the stability region when both eigenvalues are between the black limits $-1 < \lambda_1(a),\lambda_2(a) < 1$

enter image description here


Where does the $1$ in your conditions come from? It is supposed to be a $0$, as far as I know. The eigenvalues show "where your system tends to" after a slight disturbence of the equilibrium. So positive ( = growing) or negative ( = decaying) is a sensible property to discuss.

To get an idea what is supposed to happen, you can in a 2x2 system always use that $$-\dfrac{1}{a} = \mathrm{det}(A)=\lambda_1\cdot\lambda_2,$$ $$-\dfrac{1}{a} = \mathrm{trace}(A)=\lambda_1 + \lambda_2.$$

From the first equation, you see that $\lambda_1$ and $\lambda_2$ have to have different signs. Consquently, $0$ is always a saddle point.


If you eliminate $x$, you get a second order recurrence on $y$, with the characteristic polynomial


The roots are


They are smaller than one in module, except for the "minus" one before $a=2$.


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.