Variance of lattice point counting in some special shells in \(\mathbb{R}^d\) (Q2043476)

The classical lattice point problem associated with a compact convex domain \(B\subseteq\mathbb{R}^{d}\) is concerned with counting the number of lattice points in the dilation of the domain \(rB\) for large values of \(r\). This problem can be approached by taking the volume of the domain as the main term and the boundary content as the error or remainder term. The primary focus is often to improve the error term associated with lattice points lying on the boundary, given for \(r\geq 1\) by \[ R_{B}(r)=\#(rB\cap \mathbb{Z}^{d})-\mathrm{v}\mathrm{o}\mathrm{l}(B)r^{d}. \] Variations on this approach include counting the lattice points lying within a short distance of the boundary, as considered by the reviewer [Acta Arith. 141, No. 1, 73--101 (2010; Zbl 1219.11148)]. More recently, for \(B\subseteq \mathbb{R}^{d}\) a convex domain containing the origin with smooth boundary and nonzero Gaussian curvature, it was shown by Colzani, Gariboldi and Gigante [\textit{L. Colzani} et al., J. Geom. Anal. 31, No. 9, 8903--8924 (2021; Zbl 1468.60018)] that for the shell of width \(t\), generated by the domain \(B\), notated as \[ B(r,\ t)=(r+t/2)B\backslash (r-t/2)B, \] with the resulting number of lattice points in the shell as \[ N_{B(r,t)}(u)=\#\{n\in \mathbb{Z}^{d}\ :\ n\in(B(r,\ t)-u)\}, \] that the variance of \(N_{B(r,t)}(u)\) is asymptotic to \(\mathrm{v}\mathrm{o}\mathrm{l}(B(r,\ t))\) as \(r\) goes to infinity. Here \(0<t\leq r^{-\alpha}\) for every \(\alpha>(d- 1)/(d+1)\), and \(u\) is uniformly distributed in \(\mathbb{T}^{d}\). In this paper, the authors study the variance of the random variable \[ N_{D(r,t)}(u)=\#\{n\in \mathbb{Z}^{d}\ :\ n\in(\mathcal{D}(r,\ t)-u)\}, \] where the shell \[ \mathcal{D}(r,\ t)=(r+t/2)\mathcal{D}\backslash (r-t/2)\mathcal{D} \] is generated by the convex domain \[ \mathcal{D}=\left\{x\in \mathbb{R}^{d}:\sum_{p=0}^{n-1}\left(\sum_{l=1+d_{p}}^{d_{p+1}}x_{l}^{\omega_{l}}\right)^{m_{p+1}}\leq 1\right \}. \] Here \(n,p,m_{p+1},d_{p+1}\in \mathbb{N}\cup\{0\},\ 0=d_{0}<d_{1}<.\text{ . . }<d_{n-1}<d_n=d\geq 3\), and \(4\leq \omega_{l}\in 2\mathbb{N}\). Their main result concerns the variance of \(N_{D(r,t)}(u)\) stated below. Theorem. For the domain \(\mathcal{D}(r,\ t)\), if \(1<r<+\infty, \,\,0<t\leq Cr^{-\alpha}\) for some constant \(C>0\) and \(\displaystyle \alpha>\max\{\alpha_{1},\ \alpha_{2},\ .\ .\ .\ ,\ \alpha_{d-1}\}\), then there is a constant \(\beta>0\) such that \begin{center} \(\displaystyle \int_{\mathbb{T}^{d}}|N_{D(r,t)}(u)-\mathrm{v}\mathrm{o}\mathrm{l}(\mathcal{D}(r,\ t))|^{2}\mathrm{d}u=\mathrm{v}\mathrm{o}\mathrm{l}(\mathcal{D}(r,\ t))(1+O(t^{\beta}))\) , \end{center} where the implicit constant is independent of \(r\) and \(t.\)