MATHEMATICS OF COMPUTATION, VOLUME 28, NUMBER 126, APRIL 1974, PAGES 465-473
Approximation by Aliasing with Application to
"Certaine" Stiff Differential Equations
By Arthur David Snider and Gary Charles Fleming
Abstract. The usual method of finding an accurate trigonometric interpolation for a
function with dominant high frequencies requires a large number of calculations. This
paper shows how aliasing can be used to achieve a great reduction in the computations in
cases when the high frequencies are known beforehand. The technique is applied to stiff dif-
ferential equations, extending the applicability of the method of Certaine to systems with
oscillatory forcing functions.
1. Introduction. In general, when one wishes to perform a Fourier analysis
on a periodic function fit) using sampled data, the estimation of the /Vth Fourier
coefficient requires at least IN data points ([1], [2]). The present paper shows that it
is possible to do this with a much smaller data set under special circumstances,
namely, when fit) is a sum of a smooth function and of a few harmonics of high,
known frequencies. The technique involves the use of aliasing [1] to orthogonally
project out the high coefficients with just a few computations.
This can be used to extend the applicability of Certaine's method ([3], [4]) in
numerically solving systems of stiff differential equations of the form
dy/dx = My + giy, x).
Here x is the independent variable, y and g are vector functions, and M is a matrix
with large eigenvalues. This latter property ("stiffness") will dictate the use of an
extremely fine mesh, resulting in an expensive computation, unless some special
technique is used. In [3] and [4], an integrating factor exp(Mr) is introduced to over-
come this difficulty, and the function g is approximated by interpolating polynomials,
yielding a stable, accurate predictor-corrector scheme at reasonable mesh lengths
in those cases when g is known to be smooth and slowly-varying. The trigonometric
interpolation scheme which we describe herein permits an extension of this technique
to cases where g is oscillatory, without destroying its basic attractive feature—its
employment of reasonably-sized mesh lengths.
2. The Approximation. Let us suppose that fit) has period lr, and that we
wish to estimate the Fourier coefficients for the terms sin nt and cos nt for, say, n
up to 1000, using sampled data. Normally, we would proceed as in [1]; to find a
trigonometric sum of the form
A A'~1 A
(0 CNit) = -7T + Z (Ar cos rt + Br sin rt) + -f cos Nt
Z r-l -¿
Received February 16, 1973.
AMS (MOS)subjectclassifications (1970). Primary42A08, 42A12, 65L05.
Copyright © 1974, American Mathematical Society
465
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use