Exponential sum approximations for $t^{-\beta}$

TL;DR

This paper reviews and improves exponential sum approximation methods for the function t^(-β), demonstrating enhanced initial accuracy with a new integral representation before applying Prony's method.

Contribution

The paper introduces an alternative integral representation that yields better initial approximations for exponential sums approximating t^(-β).

Findings

New integral representation improves initial approximation accuracy.

Both methods perform similarly after applying Prony's method.

Enhanced approximation efficiency for t^(-β) in compact intervals.

Abstract

Given $\beta>0$ and $\delta>0$, the function $t^{-\beta}$ may be approximated for $t$ in a compact interval $[\delta,T]$ by a sum of terms of the form $we^{-at}$, with parameters $w>0$ and $a>0$. One such an approximation, studied by Beylkin and Monz\'on, is obtained by applying the trapezoidal rule to an integral representation of $t^{-\beta}$, after which Prony's method is applied to reduce the number of terms in the sum with essentially no loss of accuracy. We review this method, and then describe a similar approach based on an alternative integral representation. The main difference is that the new approach achieves much better results before the application of Prony's method; after applying Prony's method the performance of both is much the same.