Linear independence of time frequency translates for special configurations (Q2275732)

The Heil-Ramanathan-Topiwala conjecture [\textit{C. Heil, J. Ramanathan} and \textit{P. Topiwala}, ``Linear independence of time-frequency translates'', Proc. Am. Math. Soc. 124, No. 9, 2787--2795 (1996; Zbl 0859.42023)] states that if \(\{(t_j,\xi_j)\}_{j=1}^n\) are \(n\geq 2\) distinct points in the time-frequency plane then there is no nontrivial \(L^2\) function \(f:\mathbb{R}\to\mathbb{C}\) and coefficients \(\{d_j\}\) such that \(\sum_{j=1}^n d_j \, f(x+t_j)\, e^{2\pi i \xi_j x}=0\) a.e. on \(\mathbb{R}\). The conjecture has been proved when the points \(\{(t_j,\xi_j)\}_{j=1}^n\) sit on a lattice but there is no other general condition on the points under which the conjecture was previously verified. Here, special \((p,q)\) configurations are considered under which there exist two distinct parallel lines in the plane such that one of them contains \(p\) points and the other contains the remaining \(n-p\) points. The points need not lie on a grid. The conjecture is proved for \((1,3)\) and \((2,2)\) configurations (\(n=4\)). The \(n=3\) case follows from the result when the points lie on a grid. The approach is number-theoretic. In particular, first a special case is proved in which the points on each of the lines (that contain more than one point) are, in a sense, nearly rational multiples of one another. The general case then follows by reducing to these special cases using metaplectic transformations. The \((3,1)\) case is harder than the (2,2) case. It is based on the idea that any linear dependence between time-frequency translates gives rise to a recurrence along \(\mathbb{Z}\)-orbits, \(x+\mathbb{Z}\). Diophantine approximation is then used to identify appropriate scales on which recurrence along finite portions of distinct orbits gives the desired independence by making use of almost periodicity of trigonometric polynomials.