Approximate construction of rational approximations and the effect of error autocorrection. Applications

Grigori Litvinov
DOI: https://doi.org/10.48550/arXiv.math/0101042
2001-01-05
Numerical Analysis
Abstract:Several construction methods for rational approximations to functions of one real variable are described in the present paper; the computational results that characterize the comparative accuracy of these methods are presented; an effect of error autocorrection is considered. This effect occurs in efficient methods of rational approximation (e.g., Pade approximations, linear and nonlinear Pade-Chebyshev approximations) where very significant errors in the coefficients do not affect the accuracy of the approximation. The matter of import is that the errors in the numerator and the denominator of a fractional rational approximant compensate each other. This effect is related to the fact that the errors in the coefficients of a rational approximant are not distributed in an arbitrary way but form the coefficients of a new approximant to the approximated function. Understanding of the error autocorrection mechanism allows to decrease this error by varying the approximation procedure depending on the form of the approximant. Some applications are described in the paper. In particular, a method of implementation of basic calculations on decimal computers that uses the technique of rational approximations is described in the Appendix. To a considerable extent the paper is a survey and the exposition is as elementary as possible.
What problem does this paper attempt to address?