Tight error bounds for rank-1 lattice sampling in spaces of hybrid mixed smoothness (Q2407468)

Error bounds for the reconstruction of multivariate periodic functions sampled on rank-1 lattices \(\Lambda(\pmb{z},M)\), with generating vector \(\pmb{z} \in \mathbb{Z}^d\) and lattice size \(M\), are developed. The weighted Hilbert spaces \(\mathcal{H}^{\alpha,\beta}(\mathbb{T}^d)\) of complex functions \(f:\mathbb{T}^d\rightarrow \mathbb{C}\), defined on the \(d\)-dimensional torus \(\mathbb{T}^d\), depend on smoothness parameters \(\alpha,\beta \in \mathbb{R}\). Specific errors \(g_M^{\mathrm{latt}_1}(\mathcal{F},Y)\) for functions of the class \(\mathcal{F}\) and the operator norm in the functions class \(Y\) are introduced. For functions \(f\in\mathcal{F}\), reconstructed by an operator \(A:\mathbb{C}^M\rightarrow Y\) from its samples on the lattice \(\Lambda(\pmb{z},M)=\{\mathbf{x}^1,\dots,\mathbf{x}^M\}\subset \mathbb{T}^d\), the errors are defined as \[ g^{\mathrm{latt}_1}_M({\mathcal{F}},Y)=\inf_{\pmb{z}\in\mathbb{Z}^d}\text{Samp}_{\Lambda(\pmb{z},M)}({\mathcal{F}},Y), \] with the quantity \(\text{Samp}_{\Lambda(\pmb{z},M)}\) given as \[ \text{Samp}_{\Lambda(\pmb{z},M)}({\mathcal{F}},Y)=\inf_{ A:\mathbb{C}^M\to Y}\sup_{\|f|{\mathcal{F}}\|\leq 1} \left\|f-A\big(f(\mathbf{x}^i)\big)_{i=1}^M\right\|_Y\,. \] Lower and upper bounds for various combinations of the spaces \(\mathcal{F}\) and \(Y\) are derived, notably the following bounds for \(M\in \mathbb{N}\) are stated: \[ 2^{-\frac{\alpha+1}{2}} M^{-\frac{\alpha}{2}} \leq g^{\mathrm{latt}_1}_M(\mathcal{H}^{\alpha,0}(\mathbb{T}^d),\qquad L_2(\mathbb{T}^d)) \lesssim M^{-\frac{\alpha}{2}}(\log M)^{\frac{d-2}{2}\alpha+\frac{d-1}{2}}. \] Upper bounds for the special case \(d=2\) are further developed. The results are numerically tested on a specific model function \(f\) and for a specific fast approximate reconstruction algorithm.