Define the (centered) Hardy-Littlewood maximal function by $$\mathcal{M}f(x)=\sup_{r>0}\dfrac{1}{m(B(x,r))}\int_{B(x,r)}\left|f(y)\right|dy,\ f\in L_{\mathrm{loc}}^{1}(\mathbb{R}^{d})$$

We say that an operator $T:X\rightarrow Y$ between function spaces is sublinear if $$\left|T(\lambda f)\right|=\left|\lambda\right|\left|T(f)\right| \text{ and }\left|T(f+g)\right|\leq \left|T(f)\right|+\left|T(g)\right|$$ for all $f,g\in X$ and $\lambda\in\mathbb{C}$. It is easy to see that if $T(f)\geq 0$ for all $f\in X$, then $$\left|T(f)-T(g)\right|\leq T(f-g)$$

It follows from the triangle inequality that $\mathcal{M}$ satisfies the above condition and therefore is sublinear on $L^{1}(\mathbb{R}^{d})$. Moreover, the Hardy-Littlewood-Wiener theorem tells us that $\mathcal{M}:L^{p}(\mathbb{R}^{d})\rightarrow L^{p}(\mathbb{R}^{d})$ is a bounded, sublinear operator for $1<p<\infty$. $\mathcal{M}$ is in fact continuous on $L^{p}(\mathbb{R}^{d})$. Indeed, it follows from $\left|\mathcal{M}f-\mathcal{M}g\right|\leq\mathcal{M}(f-g)$ that

$$\left\|\mathcal{M}f-\mathcal{M}g\right\|_{L^{p}}^{p}\leq\left\|\mathcal{M}(f-g)\right\|_{L^{p}}^{p}\leq C\left\|f-g\right\|_{L^{p}}^{p}$$

It is a well-known theorem of Kinnunen that $\mathcal{M}:W^{1,p}(\mathbb{R}^{d})\rightarrow W^{1,p}(\mathbb{R}^{d})$ is a bounded operator. In the paper [H. Luiro, "Continuity of the Maximal Operator in Sobolev Spaces"], the author claims that $\mathcal{M}$ is not sublinear. Is this a misprint, or am I missing something nontrivial? I understand why the argument for the $L^{p}$ case doesn't carry over for Sobolev space, since we also have to deal with the weak gradient, and therefore continuity is nontrivial. However, I don't see why restricting the domain affects sublinearity.


2 Answers 2


I think the author (somewhat unconventionally) uses the word "sublinearity" for the property $$\left\|\mathcal{M}f-\mathcal{M}g\right\| \leq\left\|\mathcal{M}(f-g)\right\| \tag{1}$$ which indeed depends on the norm used. As you noted, $(1)$ holds for $L^p$ norm. I suppose there is an example out there to show that $(1)$ fails for $W^{1,p}$ norm, but I do not have one.

  • 1
    $\begingroup$ Please do share if you come up with an example. $\endgroup$ Jan 3, 2015 at 22:19
  • $\begingroup$ Try asking the author since he must have had such an example when he made the claim. I'd like to have an example too, but time constraints... $\endgroup$
    – user147263
    Jan 3, 2015 at 22:21
  • $\begingroup$ I have an explanation for why $\mathcal{M}$ is not sublinear in the sense you defined. See my posted answer. $\endgroup$ Sep 17, 2015 at 16:28

After reading the Luiro's paper again more carefully, I realize that he answers why $\mathcal{M}$ is not sublinear in the sense that $$\left\|\mathcal{M}f-\mathcal{M}g\right\|_{W^{1,p}}\leq\left\|\mathcal{M}(f-g)\right\|_{W^{1,p}},\quad\forall f,g\in W^{1,p}(\mathbb{R}^{n})$$ in the concluding remark of the paper. The failure of sublinearity is a consequence of a more general result, which is that $\mathcal{M}$ has no modulus of continuity $\omega$ on $W^{1,p}(\mathbb{R}^{n})$. I.e., there does not exist a function $\omega:\mathbb{R}_{+}\rightarrow\mathbb{R}_{+}$ such that

$$\left\|\mathcal{M}f-\mathcal{M}g\right\|_{W^{1,p}}\leq\omega(\left\|f-g\right\|),\qquad\forall f,g\in W^{1,p}(\mathbb{R}^{n})$$

This follows from two properties: $\mathcal{M}$ is not Lipschitz continuous $W^{1,p}(\mathbb{R}^{n})\rightarrow W^{1,p}(\mathbb{R}^{n})$ and the scale invariance of $\mathcal{M}$.

If $\mathcal{M}$ were sublinear, then $\omega(t)=t$ would be a modulus of continuity for $\mathcal{M}$.

Theorem 1. [Korry 2002] There exists a $C_{c}^{\infty}(\mathbb{R}^{n})$ function $f$ such that $\mathcal{M}f$ does not belong to $F_{p,q}^{s}(\mathbb{R}^{n})$ for every $s>1+1/p$. In particular, $\mathcal{M}f\notin W^{2,p}(\mathbb{R}^{n})$.

The Triebel-Lizorkin space $F_{p,q}^{s}(\mathbb{R}^{n})$ is equivalent to the fractional Sobolev space $W^{s,p}$ when $q=2$. By Sobolev embedding, $f\notin W^{2,p}(\mathbb{R}^{n})$.

Lemma 2. [Gilbarg-Trudinger] Let $f\in L^{p}(\mathbb{R}^{n})$, $1<p<\infty$, and suppose that the difference quotient $\Delta_{i}^{h}f$ are uniformly bounded in $L^{p}$ norm by some constant $C>0$. Then $f\in W^{1,p}(\mathbb{R}^{n})$ and $\left\|\partial_{i}f\right\|_{L^{p}}\leq C$ for all $i$. Conversely, if $f\in W^{1,p}(\mathbb{R}^{n})$, then $\left\|\Delta_{i}^{h}f\right\|_{L^{p}}\leq\left\|\partial_{i}f\right\|_{L^{p}}$ for all $i$.

We now have the ingredients to prove the main result.

Proposition 3. $\mathcal{M}$ is not Lipschitz continuous.

Proof. Assume the contrary. For nonzero $h\in\mathbb{R}$, let $\Delta_{i}^{h}$ denote the difference quotient in the $i^{th}$ coordinate. Observe that by the translation invariance of $\mathcal{M}$, $$\left\|\Delta_{i}^{h}(\mathcal{M}f)\right\|_{1,p}=\left\|(\mathcal{M}(h^{-1}\tau^{-h}(f))-\mathcal{M}(h^{-1}f)\right\|_{1,p}\leq\left\|h^{-1}\tau^{-h}(f)-h^{-1}f\right\|_{1,p}=\left\|\Delta_{i}^{h}f\right\|_{1,p}$$ If $f\in W^{2,p}(\mathbb{R}^{n})$, then by Lemma 2, there exists $C>0$ such that $\left\|\Delta_{i}^{h}f\right\|_{1,p}\leq C$ for all $h\neq 0$ and $i$. Whence, $\mathcal{M}f\in W^{2,p}(\mathbb{R}^{n})$, which is false by Korry's result. $\Box$

Suppose that $\mathcal{M}$ has a modulus of continuity $\omega$ in the sense defined above. Then by scale invariance, we have that for all $f\neq g\in W^{1,p}$, \begin{align*} \dfrac{\left\|\mathcal{M}f-\mathcal{M}g\right\|_{1,p}}{\left\|f-g\right\|_{1,p}}&=\left\|\mathcal{M}(\left\|f-g\right\|_{1,p}^{-1}f)-\mathcal{M}(\left\|f-g\right\|^{-1}g)\right\|_{1,p}\\ &\leq\omega\left(\left\|\left\|f-g\right\|_{1,p}^{-1}f-\left\|f-g\right\|_{1,p}^{-1}g\right\|_{1,p}\right)\\ &=\omega(1), \end{align*} which is a contradiction.

  • $\begingroup$ Thanks for writing this up. I'd be nice to see a concrete (and not too complicated) example of how the $1$-Lipschitz property fails, in case it gives some insights; then again, perhaps it doesn't. $\endgroup$
    – user147263
    Sep 17, 2015 at 17:01
  • $\begingroup$ @NormalHuman: If I understand what your statement correctly, I would suggest looking at Korry's paper, particularly the proof of Theorem 2. I haven't worked out the details myself, but he constructs a function $f$ in one-dimension such that $Mf$ is not in $F_{p,q}^{s}$, with parameters above, and then uses $f$ to construct a function $\tilde{f}$ on $\mathbb{R}^{n}$, such that $M_{n}\tilde{f}$ is not in $F_{p,q}^{s}$. $\endgroup$ Sep 17, 2015 at 17:04

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.