On the uniqueness problem for Fourier series

From MaRDI portal



Abstract: In various applications the problem of separation of the original signal and the noise arises. For example, in the identification problem for discrete linear and causal systems, the original signal consists of the values of transfer function at some points in the unit disk. In this paper we discuss the problem of choosing the points in the unite disk, for which it is possible to remove the additive noise with probability one. Since the transfer function is analytic in the unite disk, so this problem is related to the uniqueness theorems for analytic functions. Here we give a new uniqueness result for bounded analytic functions and show its applications in the denoising problem.


Denote by \(S_n(x, f)\), \(n= 1,2,\dots\), the partial sums of the Fourier series of a function \(f\in L_1(-\pi,\pi)\). The following theorem is proved: Let \(0<\alpha<1\) be such that \[ \int^\pi_{-\pi} \int^\pi_0 t^{-1-\alpha}|f(x+ t)- f(x)|\,dx\,dt< \infty, \] and let \(E\) be a subset of \((-\pi,\pi)\) such that \[ \sum^\infty_{n=1} 2^{n(1-\alpha)} C_{1-\alpha}(E_n(x_0))= \infty \] for almost all points \(x_0\in (-\pi,\pi)\), where \[ E_n(x_0):= \{x\not\in E: 2^{-1-n}<|x- x_0|< 2^{-n}\}. \] Under these conditions, if \[ \lim_{n\to\infty}\, S_n(x,f)= 0\quad\text{at each point }x\not\in E, \] then \(f(x)= 0\) almost everywhere on \((-\pi,\pi)\). We recall that the \(\sigma\)-capacity of a Borel set \(E\) is defined by \[ C_\sigma(E):= \Biggl(\inf_{\mu\prec E}\,\int_E \int_E {d\mu(x)d\mu(y)\over|x- y|^\sigma}\Biggr)^{-1}, \] where \(\mu\prec E\) means that \(d\mu\) is a probability measure with support in \(E\).











This page was built for publication: On the uniqueness problem for Fourier series

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1769728)