Inequalities on time-concentrated or frequency-concentrated functions (Q2491222)

The authors prove Fourier uncertainty inequalities that trade concentration in terms of size of support on the one hand, versus variance on the other. The angular variance of a \(2\pi\)-periodic function \(f\) with Fourier series \(f\sim \sum_{-\infty}^\infty a_j e^{ i jt}\) can be defined as \[ \Delta_\theta(f)^2 =\frac{(\sum_{-\infty}^\infty | a_j| ^2)^2- | \sum_{-\infty}^\infty a_j\overline{a_{j+1}}| ^2}{| \sum_{-\infty}^\infty a_j\overline{a_{j+1}}| ^2} \] if \(\sum_{-\infty}^\infty | a_j| ^2=1\), minimizing \(\Delta_\theta\) under some constraint amounts to maximizing \(| \sum_{-\infty}^\infty a_j\overline{a_{j+1}}| \). For a \(2\pi\)-periodic function whose spectrum is \(\varepsilon\)-concentrated on a set \(S\subset \mathbb{Z}\) in the sense that \(\sum_{j\notin S} | a_j| ^2\leq \varepsilon^2 \sum_{j\in \mathbb{Z}} | a_j| ^2\), it is proved that among all \(f\) and \(N\)-element sets \(S\), the minimizer of \(\Delta_\theta(f)\) is determined by its Fourier coefficients \(\{a_j\}\) as follows: (i) \(\sum | a_j| ^2=1\), (ii) \(\sum_{j=1}^N | a_j| ^2=1-\varepsilon\), and there are numbers \(A>0\), \(B\geq 0\), \(\gamma\in (0,\pi/(N+1)]\) and \(\alpha\in (0,1)\) such that (iii) \(a_j=A\cos \gamma(j-\frac{N+1}{2})\) for \(j=0,1,\dots,N+1\) and \(a_j=B\alpha^{j-N}\) for \(j\geq N\) and \(a_j=B\alpha^{1-j}\) for \(j\leq 1\). A parallel result is proved for functions in \(L^2(\mathbb{R})\) under additional regularity constraints.