On multivariate attenuation factors (Q1895871)

This paper is concerned with discrete attenuation factors for the approximate computation of multivariate Fourier transforms. In order to construct a unique approach to the discrete and nondiscrete settings, the groups \(G :=\mathbb{R}^d/N\mathbb{Z}^d\) \((N \in\mathbb{R},\;N > 0)\) and \(G :=\mathbb{Z}^d/N\mathbb{Z}^d\) \((N\in\mathbb{Z},\;N \geq 2)\) with the normalized Haar measures \(m_G(G) = 1\) are considered. Let \(N = mn\) \((n \in\mathbb{Z}, n \geq 2)\) and \(Q_d := [0,1)^d\). Further, let \[ U:= m\mathbb{Z}^d/N\mathbb{Z}^d = \{mj : j \in\mathbb{Z}^d \cap nQ_d\} \] be a subgroup of \(G\) of order \(|u|= n^d\). The method of attenuation factors arises from the following problem: From the known restriction \(f|_U \in L^2(U)\) of \(f \in C(G)\) we want to approximate the Fourier transform \[ \widehat {f}(k) := \int_G f(x)e^{-2\pi ixk/N} dm_G(x) \quad (k \in \widehat {G}), \] where \(\widehat {G}\) is the character group of \(G\). This problem is solved by first projecting \(f\) into a shift invariant space generated by \(l \in C(G)\), \[ If(x) := \sum_{u \in U} f(u) l(x - u) \quad (x \in G) \] and then computing the Fourier transform of \(If\). The method is applied to multivariate Bernoulli functions. For the numerical evaluation of attenuation factors, the connection between box splines and Bernoulli functions is used. Again a unique introduction of discrete and non-discrete box splines is possible considering the different groups. Furthermore, it is shown that the method of attenuation factors with the interpolation operator \(If\) (satisfying \((If)|_U = f|_U)\) is in some sense best suited for the approximation of the Fourier transform of \(f\).