The Heisenberg inequality for the discrete Fourier transform. (Q1413116)

Several uncertainty principles have been formulated quantifying various forms of joint localization of a signal \(x=(x_0,\dots,x_{N-1})\) defined on \(\mathbb Z_N\) and its discrete Fourier transform \(\widehat{x}\). For example, the Donoho-Stark uncertainty principle [\textit{D. L. Donoho} and \textit{P. B. Stark}, SIAM J. Appl. Math. 49, No. 3, 906--931 (1989; Zbl 0689.42001)] quantifies joint minimal support properties of \(x\) and \(\hat{x}\). The author begins by arguing what a fully discrete version of the Heisenberg uncertainty inequality should entail, then goes on to establish an inequality having the desired properties. He introduces the position operator \(Q: x_j\mapsto \sin(2\pi j/N)\, x_j\) and momentum operator \(P\) defined by multiplication by the matrix \(p_{ij}=\delta_{j,i+1}-\delta_{j,i-1}\) (with indices defined modulo \(N\)). The desired inequality is simply \(\| Qx\| ^2\| Px\| ^2\geq | \langle Px,Qx\rangle^2| \) which, written in coordinates for the case of \(x\) having real values, takes the form \[ 8 \| Qx\| ^2\| Q\hat{x}\| ^2\geq \bigg(\sum \Big(\sin\Big(\frac{2\pi j}{N} \Big)-\sin\Big(\frac{2\pi (j+1)}{N} \Big)\Big)a_ja_{j+1}\bigg)^2. \]