Efficient algorithm for summation of some slowly convergent series (Q608522)

This is a sequel to an earlier paper by \textit{P. Wozny} and \textit{R. Novak} [Appl. Math. Comput. 215, No.\,4, 1622--1645 (2009; Zbl 1189.65013)] concerning transformations for the extraction of a limit of a sequence. Let \( s(n)\) be a sequence in a field \(\mathbf K\) (\(n\geq 0\)). The theory may initially be related to a double sequence of integers \(m,r\geq 0\), \(m,r\) being prescribed. {\parindent=6mm \begin{itemize}\item[(a)]The nonnegative integer \(i(m,r)\) partly indicates the point in \(s\) at which a transformation is to be applied, \item[(b)]\( t(m,r)\) is the effective length of a transformation, \item[(c)]\(L(a,t\| m,r)\) is the succession of \(\mathbf K\)-valued functions \(a(m,r\|0),\dots, a(m,r\|t(m,r))\) and defines a linear operator, \item[(d)]\(S(a,v;t\|m,r;i)\in\mathbf K\) is \(\left\{\sum a(m,r\| k)v(m + i (m,r) + k)\|0\leq k\leq t(m,r)\right\}\). \item[(e)]The special \(S\)-value derived from the special \(v\)-sequence for which \(v(n)=1\), \((n\geq 0)\) is denoted by \(S(a,1;t\|m,r;i)\) and \item[(f)]if, for the prescribed \(m,r\), the denominator expression has nonzero value, then \(B(a,s;t\|m,r;i)\) is the value of the barycentric transform \[ S(a,s;t\|m,r;i)/S(a,1;t\|m,r;i)\tag{\(*\)} \] in \(\mathbf K\) of the subsequence \(s(m + i (m,r) + k)\), \(0\leq k\leq t(m,r)\). \end{itemize}} Subject to suitable conditions the sequence \(s\) has a limit to which the values of the transforms \(B\) are approximations. An implicit variant of the construction in which the linear operators \(L\) are constructed from the sequence \(s\) is also feasible. With \(c(k) = s(k+1) - s(k)\), \((k\geq 0)\), the transform coefficients \( a(m,r\|k)\) have to satisfy the orthogonality condition \[ \left\{\sum a(m,r\|k).c(m+k)\|0\leq k\leq t(m,r)\right\}=0 \quad (m\geq 0)\tag{\(**\)} \] for \(r\geq 1\). When \(r=t(m,1)= 1\), this is easily contrived: one may set, for example, \(a(m,1\|0)=1/c(m)\), \(a(m,1\|1)=-1/c(m+1)\). Defining the polynomial \(q(m,r)\) in the free variable \(E\) by \[ q(m,r\|E) = \left\{\sum a(m,r\|k)\cdot E^{k}\|0\leq k\leq t(m,r)\right\}, \] a relationship of the form \(q(m,r\|E)=P(r|E)\cdot q(m,r\|E)\) can be established, \(P(r|E)\) being another polynomial. (The \(B(a,s;t\|m,r;i)\) are then defined in terms of \(s\) alone and can be expressed as \(B(s\|m,r;i)\).) For the case in which \(i(m,r)=0\) and \(t(m,r)=r\) two algorithms for determining the \(P\) are given in the earlier paper [loc. cit.], and a third one is offered in the present paper. Considerable repetition of material from the former paper is present (see the review of that paper [loc.cit]). That review is vitiated by two typographical errors due entirely to the present reviewer: the definition of a barycentric transform should be as for (\(*\)) above and an orthogonality condition as for (\(**\)).