MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $V$ be a smooth vector field on $\mathbb{R}^n$. Assume that the maximal solution to the Cauchy problem $x'=V(x), x(0)=x_0$ exist only for $t\in [0,T)$, where $T$ is finite, denote this time by $T(x_0)$. Is $T$ continuous with respect to $x_0$ ? Is it $C^1$ ?

share|cite|improve this question
up vote 3 down vote accepted

No, but you can say it is lower semicontiuous, even wrto the initial time (that is, the optimistic situation: perturbing a little the initial data the existence is ensured almost up to $T$ , and could even be much greater) .

Precisely, given a Banach space $E$, an open set $\Omega\subset \mathbb{R}\times E$ and $f:\Omega\subset E\times \mathbb{R}\rightarrow E$ in the Cauchy-Lipshitz-Picard hypoteses, for any $(t_0,x_0)\in \Omega$ the Cauchy problem $$u(t_0)=x_0$$ for the ODE $$\dot u =f( t, u(t))$$

admits a maximal solution defined in an interval $\big(\tau_*(t_0,x_0), \tau^*(t_0,x_0)\big)\subset\mathbb{R}$, where $$\tau _ *:\Omega\to [-\infty,0 ) $$ is upper semicontinuous and $$\tau ^ *:\Omega\to (0,+\infty]$$ is lower semicontinuous. This amount to saying that: the domain of the "general solution" $\xi:\Xi\subset \Omega\times\mathbb{R}\rightarrow E$ defined as $\xi(t_0,x_0,t):=u(t)$ with the solution $u(t)$ of the above Cauchy problem, that is the set $$\Xi:=\{ (s,x,t)\in \Omega\times\mathbb{R} \, :\, \tau _ * (s,x) < t < \tau ^ *(s,x) \}$$ (that is the zone between the graph of $\tau _ * $ and the graph of $\tau ^ *$), is an open set.

share|cite|improve this answer
Interesting, do you know of an example where a sequence of initial conditions $x_i$ with $\tau^*(x_i)=+\infty$ converges to an $x$ with $\tau(x)<+\infty$ ? I have some intuition with nested periodic orbits which turn faster and faster as you get to the "outside", and which finaly go to a finite-time blow-up orbit, but I don't see how to make it rigorous. – Thomas Richard Nov 28 '12 at 10:47
By the way, do you have a reference for this theorem ? – Thomas Richard Nov 28 '12 at 10:48
To the first question: one (trivial) example is: just remove a point (e.g. $\Omega:=\mathbb{R}^2\setminus\{(1,1)\}$, and $f(t,x)=1$, so that the solution with $u(0)=0$ ends at time $1$, whereas with $u(0)=\epsilon$ they last forever). But clearly there are examples of scalar ODE's with $f:\mathbb{R}^2\to\mathbb{R}$, maybe not immediate to wirte down. But there is no obstruction to having one solution blowing up at $T=1$, e.g. $u(t)=1/(1-t)$ for $t <1$ and below its graph, solutions defined for all times. – Pietro Majer Nov 28 '12 at 11:33
Reference: the somehow geometric way I told it is perhaps written e.g. in Cartan's book (Calcul Differentiel) or in books on Dynamical Systems (maybe Arnold's). But actually most textbooks on ODE should have it, maybe stated in other forms. It's part of the property of continuous dependence by initial data. – Pietro Majer Nov 28 '12 at 11:39
Ok, I think I see why it follows from the continuity of the flow of an ODE now. Thanks ! – Thomas Richard Nov 28 '12 at 12:54

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.