A directional uncertainty principle for periodic functions (Q2002419)

Direction operators \(\mathcal{A}_L f(x)=e^{2\pi i \langle L, x\rangle} f(x)\) and \(\mathcal{B}_L f(x)=\frac{i}{2\pi}\frac{\partial f}{\partial L}(x)\) are defined respectively on \(L^2(\mathbb{T}^d)\) and \(H^1(\mathbb{T}^d)\) normed by \(\sum_{k\in\mathbb{Z}^d} |k| |c_k|\) when \(f(x)=\sum c_k e^{2\pi i k\cdot x}\). The main uncertainty inequality (Theorem 3) states that for \(L\in\mathbb{Z}^d\) and \(f\in H^1(\mathbb{Z}^d)\), the product \[\mathrm{UP}_L^{\mathbb{T}^d} f=\frac{1}{\|L\|_2^4}\left(\frac{\|f\|_{\mathbb{T}^d}^4}{|\langle \mathcal{A}_Lf,\, f\rangle |^2}-1\right) \left(\frac{\|\mathcal{B}_Lf\|_{\mathbb{T}^d}^2}{\|f\|_{\mathbb{T}^d}^2}-\frac{|\langle \mathcal{B}_Lf,\, f\rangle |^2}{\|f\|_{\mathbb{T}^d}^4}\right), \] where \(\|\cdot\|_{\mathbb{T}^d}\) denotes the \(L^2\)-norm on \(\mathbb{T}^d\), which can also be expressed as a product of variances of \(\mathcal{A}_L f\) and \(\mathcal{B}_L f\), is equal to \[ \left(\frac{(\sum_{k\in\mathbb{Z}^d}|c_k|^2)^2}{|\sum_{k\in\mathbb{Z}^d} c_{k-L}\overline{c}_k|^2} -1\right) \left(\frac{\sum_{k\in\mathbb{Z}^d}\langle L,k\rangle^2|c_k|^2}{\sum_{k\in\mathbb{Z}^d} |c_k|^2} -\left(\frac{\sum_{k\in\mathbb{Z}^d}\langle L,k\rangle^2|c_k|^2}{\sum_{k\in\mathbb{Z}^d} |c_k|^2}\right)^2\right) \] and bounded below by \(1/4\). It is shown that \(\mathrm{UP}_L^{\mathbb{T}^d} f\) is invariant under shifts, modulations and multiplications by constants; that if \(p_n(x)=(1+\cos 2\pi \langle L,x\rangle)^n\) (\(n\in\mathbb{N}\)) then \(\mathrm{UP}_L^{\mathbb{T}^d} p_n=\frac{1}{4}+O\bigl(\frac{1}{n}\bigr)\) (\(n\to\infty\)); that for Dirichlet kernels \(D_n(x)=\sum_{k=-N}^N e^{2\pi i \langle k,x\rangle}\) (\(N\) a positive multi-index) one has \(\mathrm{UP}_L^{\mathbb{T}^d} D_N\to \infty\) as \(|N|\to\infty\); and that for the Fejér kernel \(F_n(x)=\sum_{k\in\mathbb{T}^d, \max\{|k_j|\}\leq n}\left(1-\frac{\max\{|k_j|\}}{n}\right) e^{2\pi i \langle k,x\rangle}\), one has \(\mathrm{UP}_L^{\mathbb{T}^d} F_n\to \frac{(d+1)^2(d+2)^2}{6d(d+3)(d+4)}\). Results are compared to prior (nondirectional) uncertainty inequalities for \(\mathbb{T}^n\) due to \textit{S. S. Goh} and \textit{T. N. T. Goodman} [Appl. Comput. Harmon. Anal. 16, No. 1, 19--43 (2004; Zbl 1036.94529)]. Finally, a result due to \textit{H. Rauhut} [Adv. Comput. Math. 22, No. 1, 1--20 (2005; Zbl 1069.42001)] is generalized. A thread in \(S\subset\mathbb{Z}^d\) is a subset \(U=\{k,k+L,\dots, k+mL\}\). Theorem 4 states that the minimal angular variance for unit norm trigonometric polynomials with coefficient support inside a set \(S\) is \(\tan^2\frac{\pi}{m_0+1}\) where \(m_0\) is the length of the longest thread in \(S\). The Fourier coefficients of the variance minimizing polynomial are \(c_k=\sin(\pi j/(m_0+2))\), \(k=k_0+(j-1)L\). As consequences of these results some well-localized multivariate periodic Parseval wavelet frames are introduced and their uncertainty products are quantified.