The restricted discrete Fourier transform

TL;DR

This paper develops a specialized basis for the discrete Fourier transform restricted to functions supported on a finite interval, enabling explicit reconstruction formulas and eigenfunction characterizations.

Contribution

It introduces a tridiagonal matrix commuting with the DFT, providing a new eigenbasis and explicit interpolation formulas for functions supported on finite intervals.

Findings

Constructed a matrix with specific commutation and eigenspace properties.

Derived explicit Fourier interpolation formulas using theta functions.

Provided formulas for eigenfunctions when the support dimension is small.

Abstract

We investigate the restriction of the discrete Fourier transform $F_N : L^2(\mathbb{Z}/N \mathbb{Z}) \to L^2(\mathbb{Z}/N \mathbb{Z})$ to the space $\mathcal C_a$ of functions with support on the discrete interval $[-a,a]$, whose transforms are supported inside the same interval. A periodically tridiagonal matrix $J$ on $L^2(\mathbb{Z}/N \mathbb{Z})$ is constructed having the three properties that it commutes with $F_N$, has eigenspaces of dimensions 1 and 2 only, and the span of its eigenspaces of dimension 1 is precisely $\mathcal C_a$. The simple eigenspaces of $J$ provide an orthonormal eigenbasis of the restriction of $F_N$ to $\mathcal C_a$. The dimension 2 eigenspaces of $J$ have canonical basis elements supported on $[-a,a]$ and its complement. These bases give an interpolation formula for reconstructing $f(x)\in L^2(\mathbb{Z}/N\mathbb{Z})$ from the values of $f(x)$ and $\widehat f(x)$ on $[-a,a]$, i.e., an explicit Fourier uniqueness pair interpolation formula. The coefficients of the interpolation formula are expressed in terms of theta functions. Lastly, we construct an explicit basis of $\mathcal C_a$ having extremal support and leverage it to obtain explicit formulas for eigenfunctions of $F_N$ in $C_a$ when $\dim \mathcal C_a \leq 4$.