4
$\begingroup$

Let $f\in L_{\text{loc.}}^1(\Bbb{R}^d)$ such that for some $p\in (0,1)$,

$$\left\vert\int f(x)\,g(x)\,\mathrm dx\right\vert\leq\left(\int\vert g(x)\vert^p\,\mathrm dx\right)^{\frac{1}{p}}$$

for all $g\in C_c^1(\Bbb{R}^d)$, i.e., all continuous functions of compact support. Then show $f=0$ a.e.

Attempt/Thoughts:

We know $f<\infty$ a.e. because it’s in $L^1$, so if

$$A_k:=\left\{x:f(x)>\frac{1}{k}\right\},$$

one would like to show for each $k$ that $m(A_k)=0$. Then I said for $k=1$, one has $A_1=\{x:f(x)>1\}$ and since $f$ may be negative,

$$\int_{\Bbb{R}^d}f(x)\,g(x)\,\mathrm dx=\int_{\Bbb{R}^d\setminus A_1}f(x)\,g(x)\,\mathrm dx+\int_{A_1}f(x)\,g(x)\,\mathrm dx.$$

Could I say, suppose $m(A_1)\neq 0$? Then since $A_1^c$ also has infinite Lebesgue measure, then does this somehow tell me $\int_{\Bbb{R}^d\setminus A_1} f(x)\,g(x)\,\mathrm dx$ is not finite? Or can I write $\Bbb{R}^d$ as (And if $B_k:=\left\{x:f(x)<-\frac{1}{k}\right\}$)

$$\bigcap_{k}A_k \cup \bigcap_k B_k\cup\left\{x:f(x)=0\right\}.$$

But we know the $A_k,B_k$ are compact since $g \in C_c^1(\Bbb{R}^d)$.

$\endgroup$
4
  • $\begingroup$ I think the fact that $p < 1$ has to be a crucial step of the argument. If $p=1$, then with the choice of $f(x)=\frac{1}{2}$, the inequality will hold for any function $g$. $\endgroup$ Commented Sep 22 at 0:53
  • $\begingroup$ @DanielP I thought so too! I hadn't even used this!! $\endgroup$ Commented Sep 22 at 0:58
  • $\begingroup$ The inequality is supposed to hold for all $g\in C_c^1(\Bbb{R}^d)$. You seem to work with one particular $g$. Some theorems on approximation of integrable functions by functions in $ C_c^1(\Bbb{R}^d)$ is required to prove this result. $\endgroup$ Commented Sep 22 at 4:44
  • $\begingroup$ By taking limits of a sequence of mollifiers, you get that the inequality holds for all functions $g \in L^{\infty}(\mathbb{R}^d)$ vanishing outside bounded subsets. So let $g(x) = e^{i \theta(x)} \chi_E(x)$ for an optimal choice of phase (to take the absolute value inside), and some bounded set $E$. You get that $\int_E |f| \, d\lambda^d < \delta^{\alpha}$ for any bounded sets $E$ with $\lambda^d(E) < \delta$, where $\alpha = p^{- 1} > 1$. Then by taking any unit-length cube $Q$, and dissecting it repeatedly to get $\sim 2^{k d}$ cubes of measure $\sim 2^{- k d}$ each, you get $f = 0$ on $Q$. $\endgroup$ Commented Sep 22 at 5:11

1 Answer 1

8
$\begingroup$

By restricting $f$ to a fixed compact $K$ we may assume that $f$ is integrable. Applying dominated convergence on both sides we get that the inequality holds for all integrable bounded $g$.

By Lebesgue's Differentiation, $$ f(t)=\lim_{r\to0}\frac1{|B_r(t)|}\int_{B_r(t)}f(x)\,dx $$ almost everywhere on $K$. For all such $t$, $$ |f(t)|=\lim_{r\to0}\frac1{|B_r(t)|}\bigg|\int_{\mathbb R^d}f(x)\,1_{B_r(t)}\,dx\bigg|\leq\lim_{r\to0}|B_r(t)|^{\frac1p-1} =c\,\lim_{r\to0}r^{d (1-p)/p}=0 $$ since the exponent is positive. So $f=0$ a.e. on $K$, and as this can be done for all $K=\overline{B_n(0)}$, $f=0$ a.e.

$\endgroup$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.