Comparison of four nonlinear transforms on some classes of logarithmic fixed point sequences (Q1903676)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Comparison of four nonlinear transforms on some classes of logarithmic fixed point sequences
scientific article

    Statements

    Comparison of four nonlinear transforms on some classes of logarithmic fixed point sequences (English)
    0 references
    1 February 1996
    0 references
    Let \(D(n) = S(n) - L\) \((n \geq 0)\) be the discrepancies between the members \(S(n)\) of a sequence and its limit \(L\), and let \(F(x) = x + \{\sum \alpha (k) x^{1 + kr}\mid (k \geq 1)\}\) for sufficiently small \(x\), where \(r \geq 1\) is an integer. If the coefficients \(\alpha(k)\) satisfy certain conditions, discrepancies for which \(D(n + 1) = F\{D(n)\}\) have, as \(n\) increases, an asymptotic representation of the form \(D(n) \sim n^{-1/r}\{\sum \beta (k|n) n^{-k+1}\mid (k \geq 1)\}\) where, in particular, \(\beta(k|n)\) is for \(k \geq 1\) a polynomial of degree at most \(k - 1\) in \(\log (n)\). Applications of various nonlinear recursive convergence acceleration algorithms (a modified \(\rho\)-algorithm, the \(\theta\)-algorithm, Lubkin's and Steffenson's methods) to sequences of the above type are studied. Asymptotic error estimates in transformed sequences are given and numerical examples are provided. The methods considered have one affliction in common: gallopping instability -- and no examination of error propagation is conducted. For this reason it might have been of interest to have included in the survey a condensation method due to \textit{A. van Wijngaarden} [Course scientific computing B; process analysis, Math. Centre CR 18, Amsterdam (1965); see also: \textit{J. W. Daniel}, Math. Comp. 23, 91-96 (1969; Zbl 0183.44101)] which is relatively unaffected by instability. (Implementation of the method as a nice exercise in the use of a recursive procedure is given by \textit{P. W. Hemker} (ed.), NUMAL: numerical procedures in Algol 60, MC Syllabus 47. 1-47.7, Math. Centre, Amsterdam (1980); for a C-implementation, see: \textit{H. T. Lau} [A numerical library in C for scientists and engineers, CRC Press (1995; Zbl 0815.65001)]; a necessarily clumsy FORTRAN implementation is given by: \textit{P. Wynn} [Numal in FORTRAN, IIMAS, Universidad Nacional Autonoma de México, Comunicaciones técnicas Nos. 48.0-48.11 (1981)]).
    0 references
    0 references
    0 references
    0 references
    0 references
    logarithmic fixed point sequences
    0 references
    nonlinear sequence transformations
    0 references
    convergence acceleration
    0 references
    asymptotic error estimates
    0 references
    Lubkin's method
    0 references
    nonlinear recursive convergence acceleration algorithms
    0 references
    Steffenson's methods
    0 references
    numerical examples
    0 references
    instability
    0 references
    error propagation
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references