Atomic norm minimization for decomposition into complex exponentials and optimal transport in Fourier domain (Q2202787)

In this paper, atomic decompositions of complex vectors \(v = (v_m)_{m=-M}^M\) are studied. An atom is a vector \(a(z) = \big(\exp(- {\mathrm i}\,m\,{\mathrm{arg}}\,z)\big)_{m=-M}^M\) with \(z \in {\mathbb T} = \{z \in \mathbb C : \,|z|=1\}\). If the vector \(v\) has the atomic decomposition \[ v = \sum_{k=1}^K c_k \,a(z_k) \] with distinct \(z_k \in \mathbb T\) and \(c_k \in {\mathbb R} \setminus \{0\}\), then the discrete measure \[ \mu = \sum_{k=1}^K c_k \,\delta(z_k) \] explains \(v\), where \(\delta(z_k)\) is the Dirac measure. The author reviews properties of atomic decompositions, the close relationship with positive semidefinite Toeplitz matrices, and the Prony method to extract the parameters from a linear combination of atoms. He presents a characterization of the existence and uniqueness of a measure explaining \(v\) with minimal total variation norm. This minimal measure is numerically determined by a convex semidefinite program and Prony's method. Further, the author explores the optimal transport between two measures, of which only a finite sequence of Fourier coefficients is known.