Let $f: \Omega = \mathbb R \times \mathbb R^n \to \mathbb R^n$ continuous with $f(t,x) = f(x)$, locally Lipschitz such that $|f| \leq M$ in $\Omega$. Show that for every $t \in \mathbb R$ $$\begin{align}\varphi_t : \mathbb R^n &\to \mathbb R^n\\x_0 &\mapsto \varphi(t, x_0)\end{align}$$ is a homeomorphism, where $\varphi (t,x_0)$ is a unique solution defined in $\mathbb R$ for the IPV $$x' = f(x) ,\,\, x(0) = x_0 \tag{*}$$


  • $\varphi_t$ is injective

As $\varphi (t,x_0)$ is unique for each $x_0$, if $x_0 \neq y_0$ then $\varphi_t (x_0) = \varphi (t,x_0) \neq \varphi (t,y_0) = \varphi_t (y_0) $.

  • $\varphi_t$ is surjective

$\varphi_t$ is injective and defined from $\mathbb R^n$ to $\mathbb R^n$

  • $\varphi_t$ is continuous

Suppose it isn't. Then there exists a sequence of points $x_n$ in $\mathbb R^n$ such that $x_n \to x_0$ and $\varphi (t,x_n)\not\to \varphi(t,x_0)$. Consider $\varphi_n(\tau) = \varphi(\tau,x_n)$, where, $\tau \in [0,t]$. Then

$$|\varphi'_n(\tau)| = |\varphi'(\tau, x_n)| = |f(\tau, \varphi(\tau, x_n))| \leq M$$

Then the set $E =\{\varphi_n : [0,1] \to \mathbb R^n\}$ is equicontinous. Plus, by the Mean Value Theorem

$$|\varphi_n(\tau) - \varphi_n (\sigma)| \leq \sup_{\theta \in [0,1]} |\varphi'(\theta)| |\tau - \sigma| \leq M |\tau - \sigma|$$

Thus $(\varphi_n)$ is uniformly bounded. By the Arzelá-Ascoli Theorem there exists a subsequence $(\varphi_{n_k})$ converging to $\hat\varphi_0 \neq \varphi_0$ given by

$$\varphi_{n_k} = x_{n_k} + \int_0^{\tau} f(s, \varphi_{n_k}(s)) ds$$

Letting $k\to \infty$ we have

$$\hat\varphi_{0} = x_{0} + \int_0^{\tau} f(s, \hat\varphi_{0}(s)) ds$$

which is solution to $(*)$ and $\hat\varphi_0 \neq \varphi_0$, yielding a contradiction.

  • $\varphi_t^{-1}$ is continuous

  • $\phi_{t + s} = \phi_t \circ \phi_s$, for any given $t,s \in \mathbb R$.

Note: We still don't have the Continous Dependence on solutions Theorem

I'm having trouble with the last two parts.

  • 1
    $\begingroup$ The penultimate point follows from the last. Note that $\varphi_0 = \operatorname{id}_{\mathbb{R}^n}$. $\endgroup$ – Daniel Fischer Sep 16 '15 at 21:30
  • $\begingroup$ So, is it $$\varphi_t^{-1} (\varphi (t, x_0)) = id_{\mathbb R^n} = \varphi_0$$ $\endgroup$ – Aaron Maroja Sep 16 '15 at 21:40
  • $\begingroup$ Using the last bullet point, what is $\varphi_t^{-1}$? $\endgroup$ – Daniel Fischer Sep 16 '15 at 21:41
  • $\begingroup$ Continuous. If what have done on the last bullet is correct. $\endgroup$ – Aaron Maroja Sep 16 '15 at 21:42
  • $\begingroup$ No, I mean, can you give a different expression for $\varphi_t^{-1}$ using the last bullet point and $\varphi_0 = \operatorname{id}$? $\endgroup$ – Daniel Fischer Sep 16 '15 at 21:44

There are two points you need to spend some work showing,

  1. for all $t\in \mathbb{R}$, the map $\varphi_t$ is continuous, and
  2. for all $s,t\in \mathbb{R}$, we have $\varphi_{t+s} = \varphi_t \circ \varphi_s$.

Together with the obvious (sorry) $\varphi_0 = \operatorname{id}_{\mathbb{R}^n}$, all the other points follow automatically.

The property 2. shows that for all $t\in \mathbb{R}$, we have

$$\varphi_t \circ \varphi_{-t} = \varphi_{t + (-t)} = \varphi_0 = \varphi_{-t} \circ \varphi_t,$$

thus $\varphi_t$ is a bijection whose inverse is $\varphi_{-t}$.

Hence, if property 1. is also shown, it follows that $\varphi_t^{-1}$ is continuous, and hence $\varphi_t$ a homeomorphism, since $\varphi_t^{-1} = \varphi_{-t}$ is also a map whose continuity is asserted in 1.

You have already shown the continuity of $\varphi_t$, but strictly speaking only for $t \geqslant 0$. For the case $t < 0$, the same argument works, but needs some small changes in notation. I would, however, prefer a direct argument rather than your proof by contradiction, but that's a matter of personal taste. The essence of the argument is in any case that the uniform limit of the $\varphi_{n_k}$ is again a solution of the differential equation $x' = f(x)$ - or its equivalent integral form - and so the convergence $\varphi_{n_k}(0) \to x_0$ shows that $\varphi(t,x_0) = \lim \varphi_{n_k}(t)$.

So it remains to tackle property 2. Fix an arbitrary $x_0 \in \mathbb{R}^n$, and $s,t \in \mathbb{R}$. Let $\psi \colon \mathbb{R} \to \mathbb{R}^n$ be the solution to the IVP $(\ast)$. (From the question text, it seems you already know that it has a unique solution that is defined on all of $\mathbb{R}$.)

By definition, we have

  • $\varphi_s(x_0) = \psi(s)$, and
  • $\varphi_{t+s}(x_0) = \psi(t+s)$.

But since the differential equation is autonomous, the function

$$\eta \colon \tau \mapsto \psi(\tau+s)$$

satisfies $\eta'(\tau) = f(\eta(\tau))$, hence is a solution to the IVP $x' = f(x),\, x(0) = \eta(0) = \psi(s) = \varphi_s(x_0)$. Therefore, we have $\varphi_t(\varphi_s(x_0)) = \eta(t) = \psi(t+s)$, i.e.

$$\varphi_t(\varphi_s(x_0)) = \varphi_{t+s}(x_0).$$

Since $x_0$ was arbitrary, it follows that indeed $\varphi_{t+s} = \varphi_t \circ \varphi_s$.

  • $\begingroup$ Thanks for clarifying. Yes, showing that $(*)$ has an unique solution for all $\mathbb R$ is part $(a)$ of the problem. Also, everything on the book is done for $t \geq 0$. $\endgroup$ – Aaron Maroja Sep 16 '15 at 22:44
  • 1
    $\begingroup$ But we need the $t < 0$ case for the inverse. So even if the book only cares about $t \geqslant 0$, for this it is better to also consider negative $t$. $\endgroup$ – Daniel Fischer Sep 16 '15 at 22:53
  • $\begingroup$ I see, you're right. $\endgroup$ – Aaron Maroja Sep 16 '15 at 22:54

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.