Maximal subspace averages (Q2099090)

The article presents numerous significant advances in the study of maximal directional averages motivated mainly by the Kakeya and Nikodym conjectures. For a finite subset \( \Sigma \) of the Grassmannian \(\mathrm{Gr}(d,n)\) of \(d\)-dimensional subspaces of \(\mathbb{R}^n\) and for \( s \in S \subset (0, \infty)\), consider the maximal operators \[ \mathrm{M}_{\Sigma, \{s\}} f (x) = \sup_{\sigma \in \Sigma} \langle |f| \rangle_{s, \sigma} (x) \text{ and } \mathrm{M}_{\Sigma, S} f (x) = \sup_{s \in S} M_{\Sigma, \{s\}} f (x), \] where \(\displaystyle \langle |f| \rangle_{s, \sigma} (x) = \int_{sB_n \cap \sigma} |f(x-y)| \frac{dy}{s^d} \) and \(f \in \mathcal{C} (\mathbb{R}^n)\), \(B_n\) is the \(n\)-dimensional unit ball centered at the origin, \(sB_n\) is its concentric dilate and \(dy\) denotes the Lebesgue measure on \(\sigma\). The first two main results of this paper state the following. Theorem A. For all \(N > 0\) and \(d < n\), there holds \[ \sup_{s > 0} \; \sup_{\substack{\Sigma \subset \mathrm{Gr}(d,n)\\ \# \Sigma \leq N } } \left\| \mathrm{M}_{\Sigma, \{s\}} \right\|_{L^2 (\mathbb{R}^n)} \lesssim N^{ \frac{ n - d - 1 }{ 2 (n-d) } } (\log N)^{ \frac{n-d}{2} } , \] with an implicit numerical constant depending only upon \(d\), \(n\). This bound is sharp in terms of \(N\) when \(n = d + 1\) and sharp up to the logarithmic factor in general. Theorem B. Let \(\Sigma \subset \mathrm{Gr}(d,d+1)\) be a finite set. Then \[ \left\| \mathrm{M}_{\Sigma, (0,\infty)} : L^2 (\mathbb{R}^{d+1}) \to L^{2,\infty} (\mathbb{R}^{d+1}) \right\| \lesssim ( \log \# \Sigma )^{1/2} \] and \[ \left\| \mathrm{M}_{\Sigma, (0,\infty)} : L^p (\mathbb{R}^{d+1}) \to L^p (\mathbb{R}^{d+1}) \right\| \lesssim \begin{cases} (\log \# \Sigma )^{1/p}, &p > 2, \\ \log \# \Sigma, &p = 2. \end{cases} \] These bounds are best possible in terms of the dependence on the cardinality \(\# \Sigma\). The authors also introduce the \((d,n)\)-Nikodym maximal function by \[ \mathcal{N}_\delta f (x) = \sup_{\sigma \in \mathrm{Gr}(d,n) } \fint_{x + T_\delta (\sigma)} |f(y)| \, dy , \qquad x \in \mathbb{R}^n, \] where \(T_\delta (\sigma)\) is essentially a slight fattening of the unit ball \(B_d\) on \(\sigma\) (see the formal definition in Section 2.3). They then obtain: Theorem C. Let \(\delta > 0\). There holds \[ \left\| \mathcal{N}_\delta : L^2 (\mathbb{R}^{n}) \to L^{2,\infty} (\mathbb{R}^{n}) \right\| \lesssim \begin{cases} \delta^{ - \frac{ n - d - 1 }{ 2 } }, &1 \leq d < n - 1, \\ (\log \delta^{-1} )^{1/2}, &d = n - 1 , \end{cases} \] and \[ \left\| \mathcal{N}_\delta : L^2 (\mathbb{R}^{n}) \to L^2 (\mathbb{R}^{n}) \right\| \lesssim \begin{cases} \delta^{ - \frac{ n - d - 1 }{ 2 } } (\log \delta^{-1} )^{1/2}, &1 \leq d < n - 1, \\ \log \delta^{-1} , &d = n - 1. \end{cases} \] The weak \((2,2)\) bound is best possible while the strong \((2,2)\) bound is best possible up to the logarithmic factor \((\log \delta^{-1} )^{1/2}\). The study of one-dimensional directional averages in \(\mathbb{R}^n\), corresponding to \(\mathrm{Gr}(1,n)\), is connected to the problem of determining the Hausdorff dimension of Kakeya sets in \(\mathbb{R}^n\). There is a classification of such questions for all \(d < n\), with \(\mathrm{Gr}(d,n)\) corresponding to the problem of studying the existence of \((d,n)\)-Besicovitch sets. On the other hand, suitable \(L^p\)-bounds for \(\mathcal{N}_\delta\) imply corresponding lower bounds for the Hausdorff dimension of \((d,n)\)-Nikodym sets. The article also contains several relevant results and applications that complement the theorems presented above. At the same time, the estimates used throughout the paper are based on Fourier analytic almost orthogonality principles, combined with polynomial partitioning and spatial analysis techniques based on the precise calculation of intersections of \(d\)-dimensional plates in \(\mathbb{R}^n\).