MATHEMATICSof computation, VOLUME 24, NUMBER 112, OCTOBER 1970 Two Simple Algorithms for Discrete Rational Approximation By I. Barrodale and J. C. Mason Abstract. This paper reports on computational experience with algorithms due to Loeb and Appel for rational approximation on discrete point sets. Following a brief review of the linear discrete approximation problem, the two rational algorithms are stated in a general setting. Finally, several numerical examples of applications to lu h, and 4» approxi- mation are supplied and discussed. 1. Introduction. Rational functions can often provide very satisfactory approxi- mations to discrete data. However, as with most other nonlinear approximating functions, effective algorithms that produce best discrete rational approximations are few in number and are often complicated or time-consuming. The purpose of this paper is to give further exposure to two rational algorithms due to Loeb and Appel, to extend their applicability to each of the three norms /,, l2, and /„, and to test their effectiveness on a variety of problems. Both methods are simple in the sense that they employ only a linear approximation algorithm and possibly a straight- forward iteration. For the sake of completeness, the remainder of this introductory section consists of some remarks on the general problem of best approximation on a discrete point set. Given a set X = {x,, x2, • • • , xN} of real numbers and a function j(x) defined on X, we choose an approximating function F(A, x) and select a particular form F(A*, x) which approximates f(x) satisfactorily on X, according to some criterion. Here, A = {alt a2, •• • , an) is a set of free parameters, and F(A, x) is a linear ap- proximating function only if it depends linearly upon these parameters. Thus, a rational function F(A, x) = (al + a2x)/(l + a3x) is nonlinear, and the most general linear function is F(A, x) = 2"-i û>0<(*)> where the (p,(x)'s are given linearly in- dependent functions defined on X. F(A*, x) is called a best approximation in a norm ||-|| if, for all choices of A, \\j(x) - F(A*, x)\\ á \\f(x) - F(A, x)\\. The three norms used in practice are: N /,: \\w(x)\j(x) - F(A,x)]\\1 = £ w(Xi) \f(xt) - F(A, Xi)\, i-l l2: \\w(x)[f(x) - F(A,x)]\\2 = |¿ w(Xi)[f(Xi) - F(A,Xi)]2 (. i-i /»: \\w(x)[f(x) - F(A, x)]\U = max w(X<) \f(x<) - F(A,Xi)\, 1S.SAT where {w(x,)} is a prescribed set of positive weights. In the linear case best approximations exist in all three norms, but only l2 ap- Received January 16, 1970. AMS 1969 subject classifications. Primary 6520; Secondary 4117. Key words and phrases. Rational approximation, Loeb, Appel, weighted linear approximation, computational experience. Copyright© 1971, American Mathematical Society 877 | 1/2 J ' License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use