Chebyshev domain truncation is inferior to Fourier domain truncation for solving problems on an infinite interval (Q1116685): Difference between revisions

Efficient approximation by means of Fourier expansion on a subinterval (- L,L) is considered for a function u(x) defined on the real line, expandable in Fourier series, and vanishing at infinity, \(u(x)=O(1/x^ 2)\), \(x\to \infty\). The main point made in the paper is that both the number of terms N in the approximating Fourier series and the interval size L must be selected together. This is seen by dividing the function into periodic and non-periodic parts \(u(x)=P(x)+\bar P(x),\) \(P(x)=\sum^{\infty}_{n=-\infty}u(x-2nL),\) \(\bar P(x)=u(x)-P(x),\) and assessing the asymptotic behavior of the Fourier coefficients of P(x) and \(\bar P(x)\) separately. Asymptotic assessments are given assuming \(\bar P(x)\) to be real analytic and P(x) to be analytic in a horizontal strip about (-\(\infty,\infty)\) which imply that for large n the Fourier coefficients of \(\bar P(x)\) will eventually dominate those of P(x): this is shown explicitly for the functions \(u(x)=c^ 2/(c^ 2+x^ 2/L^ 2)\) and \(u(x)=\sec h(\pi x/2Lc).\) The best approximation strategy is to choose L large to make \(\bar P(x)\) small and then to adjust the number of Fourier coefficients N so that the coefficients of P(x) and \(\bar P(x)\) have relatively the same order of magnitude. Since the Fourier coefficients of P(x) decrease more rapidly than the corresponding Chebyshev polynomial coefficients the Fourier series approximation will be in general better than a Chebyshev approximation.