Multiscale matrix pencils for separable reconstruction problems (Q6145562)

The authors consider the data fitting problems: Find the \(n\) parameters \(\phi_i\) and \(\alpha_i\) such that \(f_j = \sum_{i=1}^n \alpha_i g(\phi_i;t_j)\) holds, where \(t_j\) and \(f_j\) are given. This problem is reformulated into a generalized eigenvalue problem for determining the parameters \(\phi_i\) and into a structured linear system for determining the coefficients \(\alpha_i\). At first, the case \(g(\phi_i;t_j) = \exp(\phi_it_j)\) is discussed in detail. Then, the presented approach is generalized to other types of functions such as trigonometric functions, hyperbolic functions, Chebyshev polynomials of 1st, 2nd, 3rd and 4th kind, spread polynomials, distribution functions, the \(\operatorname{sinc}\) function, and the Gamma function, where in each case special ideas for the generalization are needed. Finally, the presented methods are demonstrated by examples.