Let $\omega$ be a positive real number and let $A,B\in\mathbb{R}^{n\times n}$ be $n\times n$ real matrices. Consider the following linear time-varying dynamical system $$ \dot{x}(t) = (A + \cos(\omega t)B)x(t),\quad x(0)=x_0\in\mathbb{R}^{n}. $$

Let $\Phi(t,0)$ denote the state-transition matrix of the above system.

My question. Is it true that $$ \lim_{\omega \to \infty} \Phi(t,0) = e^{At} \ \ \ ? $$

Some remarks. If matrices $A$ and $B$ commute, this is true. Indeed, in this case it holds $$ \Phi(t,0) = e^{\int_0^t A + \cos(\omega \tau)B\, \mathrm{d}\tau}. $$ In the non-commuting case, I didn't manage to prove this. The main issue here is that a closed-form expression of $\Phi(t,0)$ does not exist, apparently. The only (perhaps useful) idea that I had so far is to exploit the Peano-Baker expansion of $\Phi(t,0)$. However, even with this tool, I couldn't provide an answer to my question.

Thanks for your help!

  • 1
    $\begingroup$ We can write the solution as $x=e^{At}y$, so we get that $y$ satisfies $y'=\cos(\omega t)C(t)y$, where $C(t)=e^{-At}Be^{At}$, and for the state-transition matrix $\Lambda(t) = e^{\int_0^t \cos(\omega \tau)C(\tau)\,d\tau}$. We see that $\Phi(t)=e^{At}\Lambda(t)$ and we must prove thus $\Lambda(t)\to I$ when $\omega\to\infty$, which seems true using integration by parts. $\endgroup$ – user90189 Jul 11 '18 at 3:58
  • 1
    $\begingroup$ @user90189: Thank you. When you have time, could you please turn your comment into an answer (perhaps elaborating a little more the different steps)? So that I will be able to accept it. $\endgroup$ – Ludwig Jul 11 '18 at 4:39
  • 1
    $\begingroup$ @user90189 I'm looking forward to your answer. I'm interested, in particular, in the proof that $y'=\cos(\omega t)C(t)y$ has state-transition matrix $\exp\bigl(\int_0^t\cos(\omega \tau)C(\tau)\,d\tau\bigr)$. $\endgroup$ – user539887 Jul 11 '18 at 8:19
  • $\begingroup$ $C(t)$ and $\int_0^t \cos(\omega s) C(s)ds$ do not necessarily commute... $\endgroup$ – copper.hat Jul 11 '18 at 14:48
  • $\begingroup$ @Ludwig, I was fooled by the form of the equation, I cannot prove that $\Lambda(t)=\exp(\int_0^t\cos(\omega \tau)C(\tau)\,d\tau)$. Sorry for misleading you. $\endgroup$ – user90189 Jul 11 '18 at 16:19

Let me turn my comment into an answer: if we write $x(t)=e^{At}y(t)$, then $y$ satisfies the differential equation $\dot{y}=\cos(\omega t)C(t)y$, where $C(t)=e^{-At}Be^{At}$. We see thus that $\Phi(t,0)=e^{At}\Lambda(t,0)$, where $\Lambda(t,0)$ is the state-transition matrix associated to $\cos(\omega t)C(t)$. Hence, we must prove that $\Lambda(t,0)y_0\to y_0$ when $\omega\to\infty$.

To show this, we integrate the differential equation to get $y(t)=y_0+\int_0^t\cos(\omega s)C(s)y(s)\,ds$, so we must prove that the integral term on the right goes to zero, but $y$ depends on $\omega$, then we write also $y_\omega$. We integrate by parts to get

$$ \frac{1}{\omega}\big[-\int_0^t\sin(\omega s)(C(s)y_\omega(s))'\,ds+\sin(\omega t)C(t)y_\omega(t)\big] \\ = \frac{1}{\omega}\big[-\int_0^t\sin(\omega s)(C'(s)y_\omega(s)+\cos(\omega s)C(s)^2y_\omega(s))\,ds+\sin(\omega t)C(t)y_\omega(t)\big]\hspace{0.5cm}(*) $$

We want to make $\omega$ goes to infinity, but we must first get a uniform upper bound of $y_\omega$. Take inner product at both sides of the differential equation to reach $\frac{1}{2}\dot{(|y|^2)}=\cos(\omega t)\langle C(t)y,y\rangle\le |C(t)||y|^2$. By Gronwall's inequality we get $|y(t)|\le R(t)|y_0|$, where $R$ is a function independent from $\omega$. Use this in equation (*), to conclude that $y(t)\to y_0$ as $\omega\to\infty$ for fixed $t$, so $\Lambda(t,0)\to I$ as $\omega\to \infty$.

  • $\begingroup$ +1: Very cute. ${}$ $\endgroup$ – copper.hat Jul 11 '18 at 23:22
  • $\begingroup$ Ingenious approach, thanks for sharing this! Out of curiosity, I was wondering if your approach can be somehow exploited to prove that for the (more general) system $\dot{x}=(A+\sum_{i=1}^m \cos(\omega_i t) B_i)x$, $\omega_i>0$, it holds $\lim_{\omega_i\to \infty}\Phi(t,0)=e^{At}$. $\endgroup$ – Ludwig Jul 12 '18 at 5:04
  • $\begingroup$ @Ludwig, I think it works. $\endgroup$ – user90189 Jul 12 '18 at 14:47

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.