I'll show my doubt with the Cauchy problem $\begin{cases} y'=1+y^2 \\ y(0)=0 \end{cases}$. I know the solution is $y(t)=tg(t)$, but let's say I don't know the explicit solution. If I look at $1+y^2$, then I have all the hypothesis for the application of Cauchy-Lipshitz Theorem in $[0, 2\pi]$ for instance, because it is lipshitz there (or not?). So I have a unique solution in $[0,2\pi]$, but now if I wonder if I can prolong the solution, how can I understand that in fact I cannot prolong the solution to $2\pi$ but at best to $\pi/2$ without solving the differential equation? I mean, where do I commit a mistake in using the Cauchy-Lipshitz Theorem? If you have some other (not simple like mine) examples that have the same problem, I would like to see how to solve them in the right way. Thanks for the help.

  • $\begingroup$ The right hand side $1+y^2$ is indeed Lipschitz continuous for $y \in [-M,M]$ for any $M$. (note that right hand side does not depend on $t$). But the Lipschitz constant goes to infinity if $M \to \infty$. Therefore the solution is indeed unique, but it could go to infinity in finite time, just as $y = \tan t$ does. $\endgroup$ – Hans Engler Mar 24 '17 at 18:40
  • $\begingroup$ Sorry, could you explain it? I don't understand why it can go to infinity finite times. I mean, the theorem states that I have a unique solution defined on all the interval (which is $\mathbb{R}$ in this case), but $tg(t)$ is not always defined. What am I missing? $\endgroup$ – tommy1996q Mar 24 '17 at 20:26
  • $\begingroup$ @tommy1996q, false. You have only local existence: en.wikipedia.org/wiki/Picard%E2%80%93Lindel%C3%B6f_theorem. $\endgroup$ – Martín-Blas Pérez Pinilla Mar 24 '17 at 21:23

Here is the statement of Cauchy-Lipschitz:

Suppose we are given an ODE, $$y'(t) = f(y(t)), \ \ \ \ \ \ \ y(0) = y_0,$$ where $f$ is a real function defined for $y \in [y_0 - b, y_0 + b]$, obeying the boundedness condition $$ |f(y)| \leq m,$$ and obeying the Lipschitz condition $$|f(y_1) - f(y_2)| \leq k |y_1 - y_2 |.$$

Then for any $$a < \min \left(\tfrac b m, \tfrac 1 k \right),$$ there exists a unique solution $y(t)$ to the ODE, valid for $t \in [-a, a]$, with $y(t)$ taking values in the range $[y_0 - b, y_0 + b]$.

Note that Cauchy-Lipschitz does NOT give us a solution for all values of $t$! It only gives us a solution for $t$ in the range $[-a, a]$, where $a$ is restricted by (i) the range of $y$ on which $f(y)$ is defined compared to how big $f(y)$ gets on this range of $y$, and by (ii) how "contracting" the function $f(y)$ is.

Let's now apply this theorem to our ODE. In our ODE, $f(y) = 1 + y^2$ and $y_0 = 0$. For the time being, let us fix a value of $b$ (so we are looking for solutions such that $y(t)$ always stays within the interval $[-b,b]$ at all values of $t$.)

On the interval $[-b,b]$, we have $$ |f(y)| \leq 1 + b^2,$$ so $|f(y)|$ is bounded by $m = 1 + b^2$. Furthermore, $$ |f'(y)| = 2|y| \leq 2b,$$ so, by the mean value theorem, we learn that $f$ is $k$-Lipschitz with $ k = 2b.$

Taking $$a < \min \left( \tfrac{b}{1 + b^2}, \tfrac 1 {2b} \right) = \begin{cases} \tfrac{b}{1 + b^2} & b \in (0, 1); \\ \tfrac 1 {2b} & b \in (1,\infty),\end{cases}$$ Cauchy-Lipschitz tells us that there exists a solution, valid for $t \in [-a, a]$, which takes values within the range $y(t) \in [-b,b]$.

The largest possible bound on $a$ is obtained when we take $b = 1$: in this case, we learn that there exists a solution valid for $t \in (-\tfrac 1 2, \tfrac 1 2)$ (and this solution takes values in the range $y(t) \in [-1,1]$, though this is less interesting).

In fact, the solution is $$ y(t) = \tan t,$$ which is valid for $t \in (- \tfrac \pi 2, \tfrac \pi 2)$. So our method of applying Cauchy-Lipschitz has given us an under-estimate for the range of $t$ on which the solution is valid.

  • $\begingroup$ I think I have understood now. Tell me if I'm wrong. The problem is that I have local existence, so once fixed $b$, that is the maximum variability of $y$ allowed, I am bound to take $a$ like that, and the interval I described couldn't be used for the theorem. Just one thing, I knew you needed local lipshitz for $y$ and uniform continuity for $t$, never heard about finiteness of $f$, is this hypothesis equivalent to the uniform continuity in this case? $\endgroup$ – tommy1996q Mar 25 '17 at 14:18
  • 1
    $\begingroup$ I have looked at the wiki page of the theorem and I think I got it now. Long story short, the proof is based on a contraction, and the values $b/m$ and $1/k$ are needed because the former makes you fall into the interval you chose for $y$, while the latter gives you a contraction. Thank you very much, you helped a lot. $\endgroup$ – tommy1996q Mar 25 '17 at 15:25
  • $\begingroup$ @tommy1996q Yes, that's exactly right! :-) $\endgroup$ – Kenny Wong Mar 25 '17 at 19:45
So I have a unique solution in $[0,2\pi]$...

False. We have only local existence:

$$\cdots$$ Then, for some value $\epsilon > 0$, there exists a unique solution $y(t)$ to the initial value problem on the interval $[t_0-\epsilon ,t_0+\epsilon]$.

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.