2
$\begingroup$

MOTIVATION.

Let $f:[0,+\infty)\to \mathbb{R}$. Solutions to $\partial_tf(t) = -\lambda f(t)$, $f(0)\neq 0$ approach zero exactly as $e^{-\lambda t}$. This property is preserved if we apply an exponentially decaying perturbation, i.e. we consider $$\partial_t f(t) = -\lambda f(t) +R(t) f(t)$$ where $|R(t)|< Ce^{-\delta t}$, $\delta >0$. To see this we can use that $f(t) = f(0) e^{-\lambda t + \int_{0}^tR(s) ds}$, and $$e^{-\lambda t + \int_{0}^tR(s) ds} \geq e^{-\lambda t + \int_0^t Ce^{-\delta s}ds} \geq e^{-\lambda t + C_1}.$$ Therefore even in the perturbed case $f(t) = \Theta(e^{-\lambda t})$ for $t\to +\infty$.

QUESTION:

what happens in higher dimension? Suppose that $f:[0,+\infty) \to H$, $H$ Hilbert space, solves $$ \partial_t f(t) = - Af + R(t)f(t)$$ $$f(0) \neq 0 $$ where $A, R(t) \in \mathcal{L}(H)$ and $||R(t)||< C e^{-\delta t}$ and $A$ is symmetric and has an (ortho)-eigenbasis $\{\psi_\lambda\}_{\lambda\in \sigma(A)}$, $\sigma(A) $ discrete, $0\not \in \sigma(A)$. Does still hold that $f_\lambda(t) = \Theta(e^{-\lambda t})$ when not $0$?

Here $f_\lambda(t) = \langle f(t), \psi_\lambda\rangle$ is the projection on the subspace generated by the eigenvector $\psi_\lambda$.

Of course if $[A,R(t)]= 0$ then the result holds as we can reduce to the 1D case.

$\endgroup$
6
  • $\begingroup$ How do you avoid the "cross terms", e.g., if in the standard basis A is a diagonal operator with eigenvalues lambda_1...lambda_n and S is the weighted unilateral shift Se_1= 0, Se_i = exp (-delta*t)e_i-1. Then your f(lambda_j+1) will depend on both lambda_j and lambda_j+1. $\endgroup$
    – Derek
    Commented Feb 4, 2022 at 14:28
  • $\begingroup$ @Derek thank you for the interest. That is exactly why it is not a trivial question. I would be happy even with some counter examples. $\endgroup$ Commented Feb 4, 2022 at 14:38
  • $\begingroup$ What does your notation $\Theta(.)$ mean? $\endgroup$ Commented Feb 4, 2022 at 14:42
  • $\begingroup$ @AlexandreEremenko Big Theta (Bachman-Landau notation en.wikipedia.org/wiki/Big_O_notation) $\endgroup$ Commented Feb 4, 2022 at 15:03
  • $\begingroup$ I think the best way to look at this problem is by using the theory of integral manifolds, see for example Aulbach and Wanner "Integral manifolds for Carathéodory type differential equations in Banach spaces", World Sci. Publ. 1996. If you have a time dependent small perturbation of a linear ODE, then you have a form of persistence, but the "eigenspaces" for the slightly perturbed spectrum also get perturbed and are time-dependent. That is, for each separable part of the spectrum, you have an invariant bundle over $t \in \mathbb{R}$. I think that when projecting on that invariant bundle, ... $\endgroup$ Commented Feb 5, 2022 at 12:01

1 Answer 1

1
$\begingroup$

$\newcommand\R{\mathbb R}$The answer is no.

E.g., suppose that $H=\R^2$, with the standard basis $(e_1,e_2)$, $Ae_1=3e_1$, $Ae_2=e_2$, $R(t)e_1=0$, $R(t)e_2=e^{-t}e_1$, and $f(0)=(0,1)$. So, $(e_1,e_2)$ is an ortho-eigenbasis of $A$, with the corresponding eigenvalues $3$ and $1$.

However, $f(t)\cdot e_1=e^{-2t}-e^{-3t}\ne\Theta(e^{-3t})$.

$\endgroup$
4
  • $\begingroup$ Do you know of any lower bound in this setting? E.g. does necessarily $f_{\lambda} (t) = \Theta(e^{-\mu t})$ for some $\mu\in \sigma(A)$ or its asymptotic can be more rapid e.g. $e^{-t^2}$? Is there a general result to your knowledge? $\endgroup$ Commented Feb 4, 2022 at 17:33
  • $\begingroup$ @WarlockofFiretopMountain : I think the lower bound $f_\lambda(t)=\Omega(e^{-\lambda t})$ will hold. I do not think that $f_\lambda(t)=\Theta(e^{-\mu t})$ for some $\mu\in\sigma(A)$ will generally hold. However, your posted question has been answered, and these or other additional questions should be posted separately. $\endgroup$ Commented Feb 4, 2022 at 17:40
  • $\begingroup$ Thanks, do not worry I always accept an answer to my first question when complete. Sometimes a follow up question is not worthy a new post, if it fits in a comment here it will fits in a comment there... Moreover, it is more natural also for occasional readers that read this question. $\endgroup$ Commented Feb 4, 2022 at 18:18
  • 1
    $\begingroup$ @WarlockofFiretopMountain : I think the lower-bound question may be interesting and nontrivial, thus deserving a separate post. $\endgroup$ Commented Feb 4, 2022 at 18:54

You must log in to answer this question.