N. Beaudoin, A high-accuracy mathematical and numerical method for Fourier transform, integral, derivatives, and polynomial splines of any order, CAN J PHYS, 76(9), 1998, pp. 659-677
From few simple and relatively well-known mathematical tools, an easily und
erstandable, though powerful, method has been devised that gives many usefu
l results about numerical functions. With mere Taylor expansions, Dirac del
ta functions and Fourier transform with its discrete counterpart, the DFT,
we can obtain, from a digitized function, its integral between any limits,
its Fourier transform without band limitations and its derivatives of any o
rder. The same method intrinsically produces polynomial splines of any orde
r and automatically generates the best possible end conditions. For a given
digitized function, procedures to determine the optimum parameters of the
method are presented. The way the method is structured makes it easy to est
imate fairly accurately the error for any result obtained. Tests conducted
on nontrivial numerical functions show that relative as well as absolute er
rors can be much smaller than 10(-100), and there is no indication that eve
n better results could not be obtained. The method works with real or compl
ex functions as well; hence, it can be used for inverse Fourier transforms
too. Implementing the method is an easy task, particularly if one uses symb
olic mathematical software to establish the formulas. Once formulas are wor
ked out, they can be efficiently implemented in a fast compiled program. Th
e method is relatively fast; comparisons between computation time for fast
Fourier transform and Fourier transform computed at different orders are pr
esented. Accuracy increases exponentially while computation time increases
quadratically with the order. So, as long as one can afford it, the trade-o
ff is beneficial. As an example, for the fifth order, computation time is o
nly ten times greater than that of the FFT while accuracy is 10(8) times be
tter. Comparisons with other methods are presented.