Polynomial-exponential decomposition from moments (Q1620891): Difference between revisions

A classical problem going back some centuries ago (see for instance [\textit{Baron Gaspard de Prony}, ``Essai expérimental et analytique: sur les lois de la dilatabilité de fluides élastiques et sur celles de la force expansive de la vapeur de l'alcool, à différentes températures'', J. École Polytechnique 1, 24--76 (1795)]) is the following: given a function \(h\in C^\infty({\mathbb R},{\mathbb C})\) of the form \[ h(x)=\sum_{i=1}^r \omega_ie^{f_ix}=\sum_{j=0}^\infty h_j\frac{x^j}{j!}, \] recover the distinct frequences \(f_1,\ldots, f_r\) and the coefficients \(\omega_1,\ldots, \omega_r\) given a finite set of ``moments'' \(h_0,\ldots, h_m\). This problem also has an ``approximation'' variant: given some moments of \textit{any} smooth function \(h(x)\) compute a representation of the form \(\sum_{i=1}^r \omega_ie^{f_ix}\) which ``best'' fit the function. The resolution proposed in [\textit{N. E. Golyandina} and \textit{K. D Usevich}, in: Matrix methods. Theory, algorithms and applications. Dedicated to the memory of Gene Golub. Based on the 2nd international conference on matrix methods and operator equations, Moscow, Russia, July 23--27, 2007. Hackensack, NJ: World Scientific. 449--473 (2010; Zbl 1216.94012)] involves the use of Hankel matrices made from the input data, and eigenvalue decomposition. The generalization of this problem to the multivariate setting is straightforward, and several approaches have been proposed to tackle it, see [\textit{F. Andersson} et al., Appl. Comput. Harmon. Anal. 29, No. 2, 198--213 (2010; Zbl 1196.42034); \textit{S. Kunis} et al., Linear Algebra Appl. 490, 31--47 (2016; Zbl 1329.65312); \textit{D. Potts} and \textit{M. Tasche}, ETNA, Electron. Trans. Numer. Anal. 40, 204--224 (2013; Zbl 1305.65093)]. Also, the study of Hankel operators of finite rank in the multivariate case has been extensively studied, see for instance [\textit{C. Gu}, Linear Algebra Appl. 288, No. 1--3, 269--281 (1999; Zbl 0947.47022); \textit{S. C. Power}, Linear Algebra Appl. 48, 237--244 (1982; Zbl 0513.15007)]. The paper under review uses duality theory for polynomial systems introduced by \textit{F. S. Macaulay} in [The algebraic theory of modular systems. With a new introduction by Paul Roberts. Reprint of the 1916 orig. Cambridge: Cambridge University Press (1994; Zbl 0802.13001; JFM 46.0167.01)] to generalize the classical challenge and tackle both multivariate problems. Via this duality, multivariate series of the form \(h(\mathbf{x})= \sum_{\alpha\in{\mathbb N}^n}\omega_\alpha\frac{\mathbf{x}^\alpha}{\alpha!}\) can be seen as elements in the dual space of an ideal of multivariate polynomials being ``anihilated'' by them. And the Hankel operator defined by it is of finite rank if and only it corresponds to ``polynomial-exponential'' series, i.e. there exist polynomials \(\omega_1(\mathbf{x}),\ldots, \omega_r(\mathbf{x})\in{\mathbb C}[x_1,\ldots, x_n]\) and \(\mathbf{\xi}_1,\ldots, \mathbf{\xi}_r\in{\mathbb C}^n\) with \(\mathbf{\xi}_i=(\xi_{i1},\ldots, \xi_{in}),\) such that \[ h(\mathbf{x}) = \sum_{i=1}^r \omega_i(\mathbf{x}) e^{\xi_{i1}x_1+\ldots\xi_{in} x_n}. \] It is shown that such polynomial-exponential series correspond -- in the algebraic language -- to Artinian Gorenstein Algebras, and to decompose them from their first moments, very fine tools from computational algebra are used to compute bases of these Algebras and perform operations in them. As by-products of these approach, Kronecker-type theorems for convolution operators and for the reconstruction of measures as weighted sums of Dirac measures from moments are provided, and a new approach for the sparse interpolation of polylog functions from values is presented.