Fall 2010 - Problem 4

Hardy-Littlewood maximal inequality, operator theory

Let TT be a linear transformation on Cc0(R)C_\mathrm{c}^0\p{\R}, the space of continuous functions of compact support, that has the following two properties:

TfLfLand{xRTf(x)>λ}fL1λ.\norm{Tf}_{L^\infty} \leq \norm{f}_{L^\infty} \quad\text{and}\quad \abs{\set{x \in \R \mid \abs{Tf\p{x}} > \lambda}} \leq \frac{\norm{f}_{L^1}}{\lambda}.

(Here A\abs{A} denotes the Lebesgue measure of the set AA.) Prove that

Tf(x)2dxCf(x)2dx\int \abs{Tf\p{x}}^2 \,\diff{x} \leq C \int \abs{f\p{x}}^2 \,\diff{x}

for all fCc0(R)f \in C_\mathrm{c}^0\p{\R} and some fixed number CC.

Solution.

First, we have the following identity:

RTf(x)2dx=R0Tf(x)2λdλdx=R02λχ{0<λ<Tf(x)}dλdx=02λRχ{0<λ<Tf(x)}dxdλ=02λm({Tf(x)>λ})dλ.\begin{aligned} \int_\R \abs{Tf\p{x}}^2 \,\diff{x} &= \int_\R \int_0^{\abs{Tf\p{x}}} 2\lambda \,\diff\lambda \,\diff{x} \\ &= \int_\R \int_0^\infty 2\lambda \chi_{\set{0 < \lambda < \abs{Tf\p{x}}}} \,\diff\lambda \,\diff{x} \\ &= \int_0^\infty 2\lambda \int_\R \chi_{\set{0 < \lambda < \abs{Tf\p{x}}}} \,\diff{x} \,\diff\lambda \\ &= \int_0^\infty 2\lambda m\p{\set{\abs{Tf\p{x}} > \lambda}} \,\diff\lambda. \end{aligned}

In the above, we switched the order of integration by applying Fubini-Tonelli, which we can do since everything is non-negative. We will first show the result for non-negative ff and extend it to a general ff. Consider the decomposition of f0f \geq 0 via a cutoff:

g=fχ{f<λ2}+λ2χ{fλ2}andh=fg.g = f\chi_{\set{f < \frac{\lambda}{2}}} + \frac{\lambda}{2}\chi_{\set{f \geq \frac{\lambda}{2}}} \quad\text{and}\quad h = f - g.

It's clear that gg is continuous on {fλ2}\set{f \neq \frac{\lambda}{2}}, since ff is continuous. For x{f(x)=λ2}x \in \set{f\p{x} = \frac{\lambda}{2}}, let {xn}nR\set{x_n}_n \subseteq \R be a sequence which converges to xx. If xn{fλ2}x_n \in \set{f \geq \frac{\lambda}{2}}, then g(xn)=λ2g\p{x_n} = \frac{\lambda}{2}. The remaining elements form a subsequence {xnk}k\set{x_{n_k}}_k and satisfies g(xnk)=f(xnk)g\p{x_{n_k}} = f\p{x_{n_k}}. By continuity of ff, g(xnk)kf(x)=λ2g\p{x_{n_k}} \xrightarrow{k\to\infty} f\p{x} = \frac{\lambda}{2}, so gg is continuous. Since hh is a difference of two continuous functions, hh is also continuous. Because ff had compact support, it follows that gg and hence hh have compact support as well.

Notice that by linearity of TT, Tf(x)Tg(x)+Th(x)\abs{Tf\p{x}} \leq \abs{Tg\p{x}} + \abs{Th\p{x}}, and so

{Tf(x)>λ}{Tg(x)>λ2}{Th(x)>λ2}.\set{\abs{Tf\p{x}} > \lambda} \subseteq \set{\abs{Tg\p{x}} > \frac{\lambda}{2}} \cup \set{\abs{Th\p{x}} > \frac{\lambda}{2}}.

Otherwise, if both Tg(x),Th(x)λ2\abs{Tg\p{x}}, \abs{Th\p{x}} \leq \frac{\lambda}{2}, then Tf(x)λ\abs{Tf\p{x}} \leq \lambda. By construction, gL=λ2\norm{g}_{L^\infty} = \frac{\lambda}{2}, so by the first property of TT, TgLTgL=λ2\norm{Tg}_{L^\infty} \leq \norm{Tg}_{L^\infty} = \frac{\lambda}{2}, and so {Tg(x)>λ2}=\set{\abs{Tg\p{x}} > \frac{\lambda}{2}} = \emptyset, which gives

{Tf(x)>λ}{Th(x)>λ2}.\set{\abs{Tf\p{x}} > \lambda} \subseteq \set{\abs{Th\p{x}} > \frac{\lambda}{2}}.

Hence,

RTf(x)2dx=02λm({Tf(x)>λ})dλ02λhL1λ/2dλ=400h(x)dxdλ=400(f(x)λ2)χ{fλ2}dxdλ=40{fλ2}f(x)λ2dxdλ40{fλ2}f(x)dxdλ=40f(x)02f(x)dλdx=80f(x)2dx.\begin{aligned} \int_\R \abs{Tf\p{x}}^2 \,\diff{x} &= \int_0^\infty 2\lambda m\p{\set{\abs{Tf\p{x}} > \lambda}} \,\diff\lambda \\ &\leq \int_0^\infty 2\lambda \cdot \frac{\norm{h}_{L^1}}{\lambda/2} \,\diff\lambda \\ &= 4 \int_0^\infty \int_0^\infty \abs{h\p{x}} \,\diff{x} \,\diff\lambda \\ &= 4 \int_0^\infty \int_0^\infty \p{f\p{x} - \frac{\lambda}{2}}\chi_{\set{f \geq \frac{\lambda}{2}}} \,\diff{x} \,\diff\lambda \\ &= 4 \int_0^\infty \int_{\set{f \geq \frac{\lambda}{2}}} f\p{x} - \frac{\lambda}{2} \,\diff{x} \,\diff\lambda \\ &\leq 4 \int_0^\infty \int_{\set{f \geq \frac{\lambda}{2}}} f\p{x} \,\diff{x} \,\diff\lambda \\ &= 4 \int_0^\infty f\p{x} \int_0^{2f\p{x}} \,\diff\lambda \,\diff{x} \\ &= 8 \int_0^\infty \abs{f\p{x}}^2 \,\diff{x}. \end{aligned}

Here, we were able to apply Fubini-Tonelli since f0f \geq 0. For a general ff, we may decompose f=f+ff = f^+ - f^- where f+=max{0,f}f^+ = \max\,\set{0, f} and f=max{0,f}f^- = \max\,\set{0, -f}, which are non-negative. Thus,

Tf(x)2dx={f0}Tf+(x)2dx+{f<0}Tf(x)2dx8f+(x)2dx+8f(x)2dx16f(x)2dx,\begin{aligned} \int \abs{Tf\p{x}}^2 \,\diff{x} &= \int_{\set{f \geq 0}} \abs{Tf^+\p{x}}^2 \,\diff{x} + \int_{\set{f < 0}} \abs{Tf^-\p{x}}^2 \,\diff{x} \\ &\leq 8\int \abs{f^+\p{x}}^2 \,\diff{x} + 8\int \abs{f^-\p{x}}^2 \,\diff{x} \\ &\leq 16 \int \abs{f\p{x}}^2 \,\diff{x}, \end{aligned}

which completes the proof.