Given $f:\mathbb{R}^n \to \mathbb{R}^n$ continuous and directionally differentiable (i.e., such that the directional derivative of $f$ exists for any direction) at a neighborhood $N$ of $x_0\in\mathbb{R}^n$, consider the ODE \begin{align*} \dot{x}=f(x) \end{align*} with initial condition $x(0)=x_0$. By Peano's existence theorem, we know that this Initial Value Problem has at least one solution. Is this solution unique? Directionally differentiable functions are not necessarily locally Lipschitz so Picard's uniqueness theorem does not apply. If uniqueness does not hold, are there any known counterexamples?
1 Answer
No. Consider $x'=g(x,y)$, $y'=h(x,y)$. If we take $g(x,y)=2|x|^{1/2}$ for $y=x^2$ similarly $h=4|x|^{3/2}$ on $y=x^2$, then we can check directly that $x=t^2$, $y=t^4$, $t\ge 0$, and $x=y=0$ are solutions with initial value $(0,0)$, independently of how we define $g,h$ off the parabola.
To define $g,h$ everywhere, we interpolate linearly, starting from the value on the parabola, by letting $g(ta,ta^2)=tg(a,a^2)$, $a\in\mathbb R$, $-1\le t\le 1$, and we also set $g(x,0)=g(0,y)=0$.
So far, $g$ has been defined on the closed set $\{ |y|\ge x^2\}\cup \{ (x,0)\}$, and it is continuous there. To verify continuity on the $y$ axis, note that if $(ta,ta^2)\to (0,b)$ with $b\not= 0$, then $|a|\to\infty$, so $tg(a,a^2)=2t|a|^{1/2}\to 0$ (and also $t|a|^{3/2}\to 0$, which we need in the corresponding step for $h$).
The existence of all directional derivatives at the origin is already guaranteed because the current domain contains a line segment in every direction, and $g$ is linear there. Finally, we extend $g$ to $\{ 0<|y|<x^2\}$. We can make $g$ smooth away from the origin.
-
$\begingroup$ +1. Do you know if there is a counterexample for $n=1$? $\endgroup$ Commented Oct 5, 2023 at 16:39
-
$\begingroup$ @IosifPinelis: A trivial remark is that when $n=1$, the assumption on $f$ is simply differentiability, so we are pretty close to a Lipschitz condition already, but that doesn't answer your question. However, there are precise criteria, as I just discovered, and it's perhaps possible to get a precise answer from this: ams.org/journals/proc/1967-018-04/S0002-9939-1967-0212240-6/… $\endgroup$ Commented Oct 5, 2023 at 17:16
-
$\begingroup$ Thank you for your response and the reference. I have looked at that paper, but it is unclear to me how to use it; in particular, how to construct such a function $V$. $\endgroup$ Commented Oct 5, 2023 at 21:13
-
$\begingroup$ Hi @ChristianRemling, thank you for your answer. However, I don't see how $g$ is continuous in the $x$ axis. The expression for $g$ in $(x,y)$ coordinates is $g(x,y)=2\frac{x^2}{|x|^{1/2}} \frac{|y|^{1/2}}{y}$. $\endgroup$ Commented Oct 5, 2023 at 21:35
-
$\begingroup$ @ToddChavez: I now think there is an easy fix. The point is that we don't really need to pay much attention to $g$ below the parabola because neither the solutions we want nor the directional derivatives depend on these data. See the edited version. $\endgroup$ Commented Oct 6, 2023 at 1:08