Let $ \Psi$ be a set of functions $\chi :\mathbb{R}^+\rightarrow [0,1)$ satisfying $$\chi(t_n)\rightarrow 1\Rightarrow t_n\rightarrow 0$$

I want to find some functions that belongs to $ \Psi$, other than

$\chi_1 (t)=\left\{\begin{matrix} e^{-2t} & \text{if } t>0\\ 1 & \text{if } t=0 \end{matrix}\right.$

$\chi_2 (t)=\left\{\begin{matrix} \frac{1}{1+t} & \text{if } t>0\\ 1 & \text{if } t=0 \end{matrix}\right.$

I'm working on this kind of functions, and i want to know more about it.

  • 1
    $\begingroup$ Well you can take a function of this form $f(t) = 0$ if $t=1 $ and $f(t)=a$ if $t\not =1$ for any $a\in(0,1)$. In this case $f(t_n)\rightarrow 0$ if and only if $t_n=0$ for all $n>N$. More generally pick any function $g(t)$ which is bounded below by some $c>0$ and change it's value in $t=1$ to be zero. $\endgroup$ – Yanko Dec 7 '18 at 14:45
  • 1
    $\begingroup$ Neither of your sample functions has the property you describe. Both satisfy $\chi(t_n) \to 0 \implies t_n \to \infty$ instead of $1$. And neither of them require a piece-wise definition, since $$e^{-2\cdot 0} = \frac 1{1 + 0} = 1$$Did you mean something else? $\endgroup$ – Paul Sinclair Dec 8 '18 at 0:47
  • $\begingroup$ @ Paul Sinclair To be honest I find it like that in a paper, and I didn't understand why the authors use piece-wise definition $\endgroup$ – Motaka Dec 8 '18 at 10:16
  • $\begingroup$ @Paul Sinclair: I'm so sorry I made an error in the definition of $\chi$.. It's correct now $\endgroup$ – Motaka Dec 10 '18 at 9:13
  • 1
    $\begingroup$ It's alright. I've modified my answer to address the corrected condition. Now I understand the piecewise definition (though it still seems a poor way to state it): $0$ is not actually in the domain of $\chi$, instead they are using this to indicate that the limit of the function at $0$ is in fact $1$. $\endgroup$ – Paul Sinclair Dec 11 '18 at 1:20

Edit: I've updated this post to match the corrected condition. As I had predicted, the updated version is very similiar to the original.

The condition $\chi(t_n) \to 1 \implies t_n \to 0$ is equivalent to (for functions $[0,\infty) \to [0,1)$):

For every $\epsilon > 0$ there is some $m_\epsilon > 0$ such that $\chi(t) \le 1- m_\epsilon$ for all $t \in [\epsilon, \infty)$.

I.e., the function is bounded away from $1$ everywhere except possibly at $t = 0$.

Your condition does not require any behavior for $\chi$ near $0$. It could be that $\lim_{t\to 0} \chi(t) = 1$, but it could also be true that $\chi$ is bounded away from $1$ everywhere. In that case, there are no sequences $\{t_n\}$ such that $\chi(t_n) \to 1$, so your condition is vacuously true.

Any function bounded away from $1$ is an example:

  • $\chi(t) = c$ for any constant $c < 1$.
  • $\chi(t) = a + b\sin t$, with $1 > a + b$ and $ a \ge b$.
  • $\chi(t) = \begin{cases} \frac{\arctan t}\pi&t \text{ rational}\\ 0.5& t\text{ irrational}\end{cases}$

Functions that satisfy the condition, but are not bounded away from $1$ near $0$ include the two you've already given, and

  • $\chi(t) = \begin{cases} 1-t & t < 1\\0& t > 1\end{cases}$
  • $\chi(t) = e^{-(t-1)^2}$
  • $\chi(t) = \frac{2\arctan (1/t)}\pi$

The boundedness condition is clearly sufficient to imply yours. To show that it is also necessary, first note that if $\chi$ is not bounded away from $1$ at some $p > 0$, then one can always find a $t_n \in (p - 1/n, p + 1/n)$ with $\chi(t_n) > 1 - 1/n$, and therefore $\chi(t_n) \to 1$ while $t_n \to p$. A similar argument holds for $p = \infty$. Thus every point in $[\epsilon, \infty]$ has a neighborhood bounded away from $1$. By compactness, a finite number of those neighborhoods covers it. The highest bound among the neighborhoods in the finite cover serves as $1-m_\epsilon$.


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.