Consider a vector space $E$ with finite dimension and linear map $A: E \rightarrow E$. The following statements are equivalent:

  • $x'(t)=A \circ x(t)$ defines an attractor.

  • All eigenvalues of $A$ have negative real part.

  • There are $\alpha, \beta, c, C>0$ and $k\geq0$ such that$$c|t^k|e^{-\alpha t}\parallel v\parallel \leq \parallel e^{tA}(v)\parallel\leq Ce^{-\beta t}\parallel v \parallel, \ \ \ v\in E$$

And, if $x'(t)=A\circ x(t)$ defines an attractor, then $$\displaystyle \lim_{t\rightarrow - \infty}\parallel e^{tA}(v) \parallel=+\infty , \ \ \ \ \ \forall v\in E$$

I have the proof for finite dimension, but I use the Jordan canonical form theorem. Clearly, I can't use this argument in infinite dimension.

Consider a linear map $A$ and $\phi :\mathbb{R}\times E \rightarrow E$ the main flow of differential equation $x'(t)=A\circ x(t)$. This equation defines an attractor when, $$\displaystyle\lim_{t\rightarrow\infty}\phi(t,v)=0$$ $\forall v \in E,$ fixed.

  • $\begingroup$ I think by "application" you mean "linear application", that is (in English) "linear map". Also, could you define "attractor"? $\endgroup$ – Lior Silberman Apr 21 '14 at 0:43
  • $\begingroup$ Now, I define an attractor. Is it clear? @LiorSilberman $\endgroup$ – Henfe Apr 21 '14 at 1:13
  • $\begingroup$ My question is in disagreement with something? $\endgroup$ – Henfe Apr 21 '14 at 2:18

As stated the Theorem is false, even for bounded self-adjoint operators on Hilbert space. Take an operator with eigenvalue $-1/k$ on the $k$th basis vector. All the spectrum has negative real parts, wich does imply $\lim_{t\to-\infty}\Vert e^{At} v \Vert = \infty$ for all (non-zero) vectors $v$, but the more precise bound above that doesn't hold -- there is no uniform exponential rate of decay.

A more precise answer will depend on the definition of "attractor".

  • $\begingroup$ Can you explain more? Can you give me an example? I do not quite understand... Thank you @LiorSilberman $\endgroup$ – Henfe Apr 30 '14 at 20:42

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.