Fix $p \in [1,\infty)$. Let $f_n:[a,b] \to \mathbb R$, $n \in \mathbb N$, be a sequence of $C^1$ functions. For every fixed $m\in \mathbb N^*$, suppose that the sequence of functions $$\{f_{n}\psi_m(f_n)\}_{n \in \mathbb N}$$ has a strongly convergent subsequence in $L^p([a,b])$ and that every subsequence $\{f_{n_k}\psi_m(f_{n_k})\}_{n_k}$ is also compact in $L^p$ for any fixed $m$. Here $\psi_m$ is a smooth cut-off function such that $0 \le \psi_m \le 1$ and $$\psi_m(f) = \begin{cases} 1 \qquad \text{ if } |f - 1|\ge 1/m \\ 0 \qquad \text{ if } |f -1 |\le 1/(2m) \end{cases} $$

Assume also that $\Vert f_n \Vert_{L^p} \le C$ (for a constant $C>0$ that does not depend on $n,m$ for all $p \in [1,\infty]$). . If necessary, also asssume that $\Vert D_x(f_n\psi_m(f_n))\Vert_{L^1} \le C_m$, where $C_m$ is a constant that depends only on $m$.

  • How can we prove (or disprove) that $\{f_n\}_{n\in \mathbb N}$ also has a strongly convergent subsequence in $L^p([a,b])$?

  • If the result is not true, what additional assumption would make it so?

This question is motivated by my previous question $L^p$ compactness for a sequence of functions from compactness of cut-off.

  • $\begingroup$ I guess the definition of the functions $\psi_m(f)$ should be something like $$\psi_m(f)(x) = \begin{cases} 1 \qquad \text{ if } |f(x) - 1|\ge 1/m \\ 0 \qquad \text{ if } |f(x) -1 |\le 1/(2m) \end{cases} $$ for each $x\in [a,b]$. $\endgroup$ – Alex Ravsky Aug 12 at 8:24
  • $\begingroup$ @AlexRavsky Yes, exactly. I was just trying to use a shorter notation. $\endgroup$ – Zac Aug 12 at 8:25
  • $\begingroup$ The condition “every subsequence $\{f_{n_k}\psi_m(f_{n_k})\}_{n_k}$ is also compact in $L^p$” is very strong. It can be shown that it is equivalent to $\{f_{n}\psi_m(f_{n})\}_{n\in\Bbb N}$ has finitely many values. I guess you mean that a set $\{f_{n}\psi_m(f_{n})\}_{n\in\Bbb N}$ is totally bounded instead. $\endgroup$ – Alex Ravsky Aug 12 at 8:30
  • 1
    $\begingroup$ @AlexRavsky Thank you. What other (more explicit) assumption on $f_n$ would make this requirement hold? $\endgroup$ – Zac Aug 12 at 9:29
  • 1
    $\begingroup$ @AlexRavsky Thanks! Is there anything that we can assume on $f_n$ (smoothness, integrability, a condition on the derivatives?) that would make this condition on $\mu(X_{n,m})$ automatically true? $\endgroup$ – Zac Aug 12 at 9:59

The question is not very clear, as stated, but the following can be proved. Let $B$ be a bounded set of $L^p(0,1)$ and assume that for every $\epsilon>0$ the set $$B_\epsilon=\{f \chi_{\{|f| \ge \epsilon\}}, \ f \in B\}$$ is relatively compact in $L^p$, then $B$ is relatively compact, too. In fact, given $\epsilon >0$, a finite $\epsilon/2$-net for $B_\epsilon$ is a $\epsilon$-net for $B$, since $(0,1)$ has finite measure. Changing characteristic functions with smooth cut-off (around 1, as in the original problem), should not change the conclusion.

| cite | improve this answer | |
  • $\begingroup$ Thank you very much for your answer. How can one prove the claim in your last sentence? $\endgroup$ – Zac Aug 11 at 21:29
  • $\begingroup$ Also, I'm not sure about the argument with nets (because I'm not very familiar with them). How can one make it more explicit? $\endgroup$ – Zac Aug 11 at 21:31
  • $\begingroup$ Sorry for the language which was a bit too cryptic. For $B$ I want to prove it is totally bounded, that is for every $\epsilon>0$ there exists a finite number of balls $B(f_i, \epsilon)$ covering it. Do it first for $B_\epsilon$ (with $\epsilon/2$ balls), by assumption, and then note that the double balls cover $B$. $\endgroup$ – Giorgio Metafune Aug 11 at 21:42
  • $\begingroup$ Ok, thanks. How does compactness follow from the covering? $\endgroup$ – Zac Aug 11 at 21:44
  • $\begingroup$ This is a standard and powerful equivalence which holds in metric spaces. $\endgroup$ – Giorgio Metafune Aug 11 at 21:49

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.