Endpoint Sobolev and BV continuity for maximal operators (Q2402684)

The \(L^{p}(\mathbb{R})\) continuity of the Hardy-Littlewood maximal function defined by \[ \tilde{M}f:x \mapsto \underset{(a,b) \ni x}{\sup} \frac{1}{b-a}\int _{a} ^{b} |f(y)|dy, \] is one of the most fundamental results in harmonic analysis. The behaviour of this operator on Sobolev spaces, however, is still far from being fully understood. One of the reasons why this question is difficult, is because \(\tilde{M}\) is not sublinear on \(W^{1,p}\), making continuity harder to prove than mere boundedness. Nevertheless, Luiro established \(W^{1,p}\) continuity for \(1<p<\infty\) in [\textit{H. Luiro}, Proc. Am. Math. Soc. 135, No. 1, 243--251 (2007; Zbl 1136.42018)]. The paper under review positively settles the harder \(p=1\) case, as well as various discrete and fractional analogues, and extensions to \(BV\) continuity. It also provides counterexamples to \(BV\) continuity of fractional maximal operators. Table 1 of the paper neatly summarises all the results, as well as the remaining open questions. The proofs start with the discrete case, where some of the authors's key ideas can be seen in a simple form. In this case, one can order the local maxima and minima of \(\tilde{M}f\) and discuss its variations between extrema. Using a Lemma from [\textit{J. Bober} et al., Proc. Am. Math. Soc. 140, No. 5, 1669--1680 (2012; Zbl 1245.42017)] that shows that \(\tilde{M}f(n) = |f(n)|\) for \(f \in BV\) and \(n\) a local maximum of \(f\), this leads to Lemma 8: a one sided control of the oscillations of the maximal function. Appropriately using this control depending on the position of the variable in the sequence of extrema, gives that \(\|f_{j}-f\|_{BV} \to 0\) implies that the \(\|\tilde{M}(f_{j})\|_{BV} \to \|\tilde{M}(f)\|_{BV}\). This, in turn, implies the result thanks to a Brezis-Lieb type reduction (Lemma 10). The continuous case is more technical, and uses a reduction to the one-sided maximal function \(M_{R}\), and a decomposition into disconnecting sets \(D_{j} = \{x \in \mathbb{R} \;;\; M_{R}f_{j}(x)>|f_{j}(x)|\}\) and connecting sets \(C_{j}= \{x \in \mathbb{R} \;;\; M_{R}f_{j}(x)=|f_{j}(x)|\}\) that replaces, to some extent, the simpler discrete idea of estimating oscillations between extrema.