4
$\begingroup$

Consider a dynamical system of the form $$ \dot{x}=f(x), \quad x\in X, $$ and assume that the system possesses a set of non-isolated fixed points. Suppose moreover that there exists a Lyapunov $V(x)$ function whose derivative is non-decreasing along the trajectories of the system. By virtue of the LaSalle invariance principle we know that the system will converge to the largest invariance set of fixed points $\{x\in X\,:\, \dot{V}(x)=0\}$. Now, I know that there exist counter-examples showing that the system can converge to some trajectories in the invariant set without necessarily approaching to a point (see for instance this post). My question is whether there exist additional conditions (different from the fact that the set of fixed points is isolated) on the system and/or Lyapunov function $V(x)$ which guarantee asymptotic converge to fixed points and not to "moving" trajectories.

Thank you in advance.

$\endgroup$
1
$\begingroup$

Yes, there do exist sufficient conditions for asymptotic stability when the Lyapunov function is negative semi-definite, which I describe below.

Krasovsky's Theorem

Given an autonomous ODE $\dot x = f(x)$ with fixed point at the origin. Let $K$ be a manifold that does not contain entire trajectories of the ODE. If there exists a Lyapunov function $V$ such that the orbital derivative $\dot V$ satisfies:

  1. $\dot V<0$ outside of $K$
  2. $\dot V=0$ on $K$

then the dynamics is asymptotically stable.

If $K = \{ x ~:~ F(x)=0\}$, here is a sufficient condition for $K$ to not contain entire trajectories of the ODE: $$ (f^T \nabla F)(x) \ne 0 \quad \text{on $K \setminus \{ \mathbf{0} \}$} $$ where $f$ is the vector field of the ODE. (This just ensures that the vector field $f$ is never orthogonal to the normal to the surface.)

Example

Consider the ODE: $ \dot x_1 = -x_1 + 3 x_2^2 \;, \dot x_2 = - x_1 x_2 - x_2^3 \;. $ In this case, a Lyapunov function is given by $V(x_1,x_2)=(x_1^2+x_2^2)/2$, whose orbital derivative is negative semi-definite. The set where $\dot V=0$ is given by the zero level set of the function $F(x_1,x_2)=x_1 - x_2^2$. Note that $(f^T \nabla F)(x_1,x_2) = 2 x_2^2 + 4 x_2^4 \ne 0$ on $K \setminus \{ \mathbf{0} \}$. Thus, by Krasovsky's Theorem the origin is asymptotically stable as illustrated in the graphic below.

graphical illustration

In this graphic, two of the axes correspond to state variables $x_1$ and $x_2$, and the other axis is the time variable $t$. The red line marks the state $x_1=0$ and $x_2=0$ for the time interval shown. Different grey shading is used for trajectories with different initial conditions.

ADD

Here is a cartoon from the book referenced below, which illustrates the idea behind Krasovsky's Theorem. The dark line labelled $\gamma$ represents a solution of the ODE, the lighter lines are contour lines of the semi-definite Lyapunov function $V$, and the dashed region is $K$ where $\dot V=0$.
cartoon This cartoon nicely illustrates how the dynamic avoids getting stuck inside $K$, and instead, asymptotically reaches a fixed point. Note that if the ODE solution enters $K$ then the value of the Lyapunov function does not change. This is illustrated by the curve remaining on a level curve of $V$ in the dashed region labelled $K$. However, eventually the ODE solution must exit $K$ (by hypothesis of Krasovky's Theorem), after which the Lyapunov function again decreases.

This picture suggests that a set of non-isolated fixed points can be reached in this fashion.

Reference

David R. Merkin [1997]. Introduction to the Theory of Stability. Texts in Applied Mathematics. Springer. Translated from Russian by Andrei L. Smirnov and Fred Afagh.

$\endgroup$
  • 1
    $\begingroup$ Thanks for the answer. In my question, I assume that the dynamical system possesses a set of non-isolated fixed points, and I ask whether there exist conditions which implies convergence to a point in the set of fixed points (and not to the set itself). It's not clear to me if your answer applies to this framework. $\endgroup$ – Ludwig Sep 22 '16 at 19:46
  • $\begingroup$ As in Lasalle's principle, the set K (for Krasovsky) is allowed to be non-isolated. Krasovsky's theorem is a general purpose result for autonomous ODEs that guarantees asymptotic convergence to fixed points and not relative equilibria, as requested. $\endgroup$ – Nawaf Bou-Rabee Sep 22 '16 at 21:19
  • 1
    $\begingroup$ I'm still a bit confused. I'm not talking about the set $K$ but the set of fixed points of $f$, namely $\mathrm{Fix}(f):=\{x\in X : f(x)=0\}$. In particular, in your example the fixed point at the origin is isolated. $\endgroup$ – Ludwig Sep 22 '16 at 23:06
  • $\begingroup$ I added the idea behind the proof. Since the proof relies on a lyapunov function and the fact that K does not contain an entire trajectory of the ODE, I think it can be adapted to your setting where the set of fixed points is non-isolated. $\endgroup$ – Nawaf Bou-Rabee Sep 23 '16 at 11:20
0
$\begingroup$

Possibly, invariant set localization tecnique can be useful to you.

Each periodic trajectory in a compact invariant set contains at least two points of the set $$ S_h= \{ x\in X : \dot h(x)=0 \}, $$ where $h(x)$ is any $C^{\infty}$ function on $X$. This fact can be used to prove non-existence of non-zero trajectories, completely contained in $\{ x\in X: \dot V(x)=0 \}$.

$\endgroup$

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.