9
$\begingroup$

Let $\Omega$ be a bounded, connected open subset of $\mathbb R^n$ with smooth boundary. For any $f \in L^1 (\Omega)$, is it true that $f \in L^\infty (\Omega)$ if and only if the following condition holds?

For every $\delta > 0$, there exists some $C > 0$ such that for almost every $x \in \Omega$, and every open ball $B \subset \Omega$ centered at $x$ with radius at least $\delta$ we have

$$|f(x)| \leq C \def\avint{\mathop{\,\rlap{-}\!\!\int}\nolimits} \avint_B |f|.$$

Note: Here $\def \avint{\mathop{\,\rlap{-}\!\int}\nolimits} \avint_B |f|$ denotes the average integral of $|f|$ over $B$.

Edit: The original question asked about the case where the balls $B$ are not necessarily centered at $x$. This has been solved in the comments - the "if" direction holds, but "only if" does not.

$\endgroup$
25
  • 2
    $\begingroup$ Concerning the "only if" part: the value of $f(x)$ is not even defined at any $x$. $\endgroup$ Oct 30, 2023 at 15:42
  • 5
    $\begingroup$ I think convex unbounded functions provide counterexamples when $d = 1$. $\endgroup$
    – Zarrax
    Oct 30, 2023 at 15:48
  • 3
    $\begingroup$ But then fix $\delta=1$ and get $|f(x)| \leq C \int_B |f| \leq C\|f\|_1$..or I misunderstood? $\endgroup$ Jan 17 at 18:23
  • 1
    $\begingroup$ @FedorPetrov Yes indeed, the “only if” direction fails horribly. Although, see my comment directly before this one. $\endgroup$
    – Nate River
    Jan 18 at 0:40
  • 1
    $\begingroup$ Yes, for centered balls there is $C$ such that this works for all Lebesgue points of $f$ $\endgroup$ Jan 18 at 6:14

2 Answers 2

8
$\begingroup$

Let me prove that for $f\in L^\infty$ and fixed $\delta>0$ (and centered balls) such $C$ exists. The idea is to find $C$ which works for all Lebesgue points of $f$. Assume that there is no such $C$, then for $C=n$ there exist violating Lebesgue points $x_n$ and $B_n$ with center $x_n$ and radius $r_n\geqslant \delta$ and $$|f(x_n)|> n\def\avint{\mathop{\,\rlap{-}\!\!\int}\nolimits} \avint_{B_n}|f|.\tag{$\heartsuit$}$$ Without loss of generality, $x_n$ converge to a point $x_0$ (it lies in $\Omega$, since all $x_n$ are at distance at least $\delta$ from $\mathbb{R}^n\setminus \Omega$, thus so does $x_0$). Then for the ball $B(x_0,\delta)$ we have $\int_{B(x_0,\delta)}f=0$: otherwise the integrals over $B_n$ would be bounded from below, and RHS's of $(\heartsuit)$ would tend to infinity, contradicting to LHS's being bounded from above by $\|f\|_\infty$. But for large $n$ we have $x_n\in B(x_0,\delta)$, and, since by $(\heartsuit)$ we have $f(x_n)\ne 0$ and $x_n$ is a Lebesgue point of $f$, we conclude that $f$ is non-zero on some set of positive measure inside $B(x_0,\delta)$. A contradiction.

For other direction, for centered balls this trivially fails for example for $f(x)=x^{-1/2}$, $\Omega=(0,1)$ (for $x$ near the singularity there are simply no centered balls of radius $\geqslant \delta$ contained in $\Omega$). For not centered balls it is However true, and settled in comments, the key word is uniform interior sphere.

$\endgroup$
5
$\begingroup$

Too long for a comment. For $d=1$ should be true. Let me simplify a bit (hopefully without loss of generality). If $\Omega=(-1,1), f\in C(-1,1)$. Then what you ask implies that \begin{equation} f(x) \leq A + B \int_0^x f(t) dt \leq A + B \int_0^x f(t)dt, \,\,\, 0\leq x <1. \end{equation} for some $A,B>0$ Then, for example by Gronwall's inequality you have \begin{equation*} f(x) \leq A e^{xB} \leq Ae^B < +\infty.\end{equation*} For $0\leq x <1$. Similarly for $-1<x\leq 0 $.

$\endgroup$
1
  • $\begingroup$ Ah, Gronwall’s inequality - very nice. $\endgroup$
    – Nate River
    Oct 30, 2023 at 17:28

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.