It is well known that the Hardy-Littlewood maximal function of a non-null $L^1(\mathbb{R})$ function doesn't belong to $L^1(\mathbb{R})$. What about the truncated Hardy-Littlewood maximal function? I.e., if $f\in L^1_{loc}(\mathbb{R})$ and $r>0$, define the $r$-truncated Hardy-Littlewood maximal function of $f$ as: $$M^*_r(f):\mathbb{R}\rightarrow[0,+\infty], x\mapsto \sup_{a\in(0,r)} \frac{1}{2a}\int_{[x-a,x+a]} |f(t)|\operatorname{d}t$$

Question 1: is it true that $M^*_r(f)\in L^1 _{loc}(\mathbb{R})?$

Question 2: if $f\in L^1(\mathbb{R})$, is it true that $M^*_r(f)\in L^1 (\mathbb{R})?$

My guess is that the answer to the first question is not true in general (and then also the answer to the second question is no in general). Anyway I have a little trouble in working out a counter-example.


If $\delta$ is a delta "function" at the origin then $M_1\delta(x)$ is like $1/|x|$ near the origin, not locally integrable. Hence $L^1$ functions that approximate $\delta$ show that there does not exist $c$ such that $$\int_0^1 M_1f\le c||f||_1.$$Not that I've worked out the details, but "surely" you can concoct an infinite linear combination of translates of functions showing that inequality fails to show that $M_1f$ need not be integrable for $f\in L^1$.

(If that's what you've been trying to do my apologies - all you told us was you were having trouble with a counterexample...)

Edit: The OP says he got it. For the benefit of anyone who hasn't got it yet: If $M_1$ were linear I'd claim were done, but it's not. Took me a minute...

For $0<\delta<1$ let $$f_\delta=\frac1\delta\chi_{(-\delta,0)}$$and $$I_\delta=(\delta,\delta^{1/2}).$$If I did the calculations correctly then $$\lim_{\delta\to0}\int_{I_\delta}M_1f_\delta=\infty.$$ One should really verify that. Assuming that's right:

Choose $\delta_k\to0$ so that the $I_{\delta_k}$ are disjoint. Let $$\alpha_k=\int_{I_{\delta_k}}M_1f_{\delta_k}.$$Since $\alpha_k\to\infty$ there exists $c_k>0$ with $\sum c_k<\infty$ and $\sum c_k\alpha_k=\infty$. Let $f=\sum c_kf_{\delta_k}$. Then $$\int_0^1M_1f\ge\sum_k c_k\int_{I_{\delta_k}}M_1f_{\delta_k}=\infty.$$

Edit${}^2$: I hate the way the above depends on the explicit calculation of $\int_{I_\delta}M_1f_\delta$. Here's a softer version, that can't possibly be wrong:

Choose a sequence of disjoint closed intervals $I_k\subset(0,1)$ with $I_k\to0$ such that $$\lim_k\int_{I_k}\frac{dx}x=\infty.$$It's clear that there exists $f_k\ge0$, supported on $(-1,0)$, with $\int f_k=1$ and $$M_1f_k(x)\ge\frac1{2x}\quad(x\in I_k).$$Now let $f=\sum c_kf_k$ as above...

| cite | improve this answer | |
  • $\begingroup$ Enough to get the idea and reach the answer, again: thanks a lot :) $\endgroup$ – Bob Aug 9 '18 at 19:55

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.