6
$\begingroup$

Consider a bump function supported in the ball of radius $1$, that is $\psi:\mathbb R^n\to\mathbb R$ such that

  1. $\ \psi(x)>0$ for $|x|<1$

  2. $\ \psi(x)=0$ for $|x|\geq 1$

  3. $\ \psi\in C^\infty$.

Is it possible to find such a function $\psi$ that satisfies also one of the following conditions? For all $i,j=1,\dots,n$

  1. $\ \displaystyle \frac{\partial_{x_i}\psi(x)}{\psi(x)}\to 0\ $, $\ \displaystyle\frac{\partial_{x_i}\partial_{x_j}\psi(x)}{\psi(x)}\to0\ $ as $|x|\to1\ $

  2. $\ \displaystyle\lim\frac{\partial_{x_i}\psi(x)}{\psi(x)}\in\mathbb R\ $, $\ \displaystyle\lim\frac{\partial_{x_i}\partial_{x_j}\psi(x)}{\psi(x)}\in\mathbb R\ $ as $|x|\to1\ $.

Intuitively this would mean that both $\psi$ and its derivatives vanish approaching the boundary of the ball, but the derivatives vanish faster than $\psi$ (or at least not slower).

A typical example of function satisfying conditions 1.-3. is given by $$ \psi_0(x) = \begin{cases} e^{-\frac{1}{1-|x|^2}} &\textrm{ if }|x|<1 \\[2pt] 0 &\textrm{ if }|x|\geq1\end{cases} $$ but 4.,5. are clearly not satisfied by $\psi_0$ since $$ \frac{\nabla\psi_0(x)}{\psi_0(x)} \,=\, \frac{2x}{(1-|x|^2)^2}\,\xrightarrow[|x|\to1]{}\,\infty \,.$$

$\endgroup$
5
  • 6
    $\begingroup$ I think the negative answer is almost implicit in the way you phrased your title - $\frac{\partial_{x_i}\psi(x)}{\psi(x)}=\partial_{x_i}(\log\psi(x))$. Since $\log\psi(x)\to-\infty$ as $|x|\to 1$, approaching the circle in the $x_i$ direction shows the limit can't be zero, or even finite. $\endgroup$
    – Wojowu
    Jul 13, 2021 at 17:06
  • 2
    $\begingroup$ By the very definition of derivative, if a smooth function vanishes at a point its first-order derivatives will, if at all, vanish always at a slower rate at the same point, so I reckon there is no such $\psi$. $\endgroup$ Jul 13, 2021 at 17:09
  • $\begingroup$ I guess you are right, thank you $\endgroup$
    – tituf
    Jul 13, 2021 at 17:11
  • 1
    $\begingroup$ @PedroLauridsenRibeiro I disagree that this is "by the very definition of derivative". The definition itself doesn't immediately imply anything about what happens if you vary $x$, so any claim on slower vanishing requires some actual argument. $\endgroup$
    – Wojowu
    Jul 13, 2021 at 17:27
  • 2
    $\begingroup$ @Wojowu sorry, bad phrasing. What I meant is that "intuitively" that should happen, considering the behavior of Newton quotients. To convert this idea into an actual proof requires work, of course. $\endgroup$ Jul 13, 2021 at 17:31

1 Answer 1

6
$\begingroup$

Elaborating the comment by Wojowu: If we take a look at $n=1$ and $\psi\in C^\infty_{\text c}(\mathbb R)$ is a function satisfying conditions 1., 2. and 3. of your question, then for every $x\in]-1,1[$, we have, by smoothness of $\ln\psi$ on $]-1,1[$ and the fact that $\frac{\psi'}{\psi}$ is continuous on every $[-K,K]$ for $K\in]0,1[$,

$$\ln\psi(x)=\ln\psi(0)+\int_0^x\frac{\psi'(s)}{\psi(s)}\,\mathrm ds.$$

Now, if we had $$\lim_{s\to 1}\frac{\psi'(s)}{\psi(s)}=r$$ for any real number $r\in\mathbb R$, then the function $$s\mapsto\frac{\psi'(s)}{\psi(s)}$$ would be well-defined and continuous on $[0,1]$, and therefore we would have $$\lim_{x\to1}\ln\psi(x)=\ln\psi(0)+\int_0^1\frac{\psi'(s)}{\psi(s)}\,\mathrm ds\in\mathbb R,$$ which is impossible since $\lim_{x\to1}\psi(x)=0$.


Similarly, for a natural number $n>1$, any $\phi\in C_{\text c}^\infty(\mathbb R^n)$ with your properties 1., 2. and 3. defines a function

$$\psi:\mathbb R\to\mathbb R, t\mapsto\phi(t,0,0,\dots, 0)$$ and we can proceed our argumentation as above.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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