Method of summation of some slowly convergent series (Q1036561)

\( i( m,r ), t( m,r ) \geq 0\) \( ( m,r \geq 0 ) \) with \(t( m,r+1 ) > t( m,r ) \) being integers, \(a(m,r | k )\) \(( m,r \geq 0;\) \( 0 \leq k \leq t( m,r ) )\) being complex numbers and \(s(\tau)\) \((\tau \geq 0)\) being a prescribed complex number sequence, set \[ S(a,s ; t \| m,r ; i ) = \{\sum a\,( m,r | k ) s\, (m + i ( m,r) + k ) | 0 \leq k \leq t\, (m,r ) \} \] for \(m,r \geq 0\), and denote the special sums obtained by taking \(s(\tau ) = 1\) \(( \tau \geq 0)\) by \(S(a,1 ; t \| m,r ; i )\). The quotient \[ B(a,s ; t \| m,r ; i ) = S\,(a,s ; t \| m,r ; i ) / S\,(a,s ; t \| m,r ; i ) \] is then a barycentric transform of the \( t( m,r ) + 1\) element subsequence \(s( m + i( m,r )) \dots s\,( m + i( m,r + t( m,r )) \); its value may, in certain cases, be nearer that of the limit or formal limit of the \(s\)-sequence than those of any of the subsequence members. If the \(a\)-numbers have been derived by imposing the condition \[ \{\sum a( m,r | k ) \{ s( m + i''( m,r ) + k) -s\,(m + i( m,r ) + k) \} | 0 \leq k \leq t\,( m,r)\} \] where \( i''(m,r ) = i\,(m,r ) + d\,(m,r )\) for some \(d( m,r ) \geq 0\), \(B( a,s ; t \| m,r ; i )\) takes the form \(B( a,s ; t \| m,r ; i'' )\) a barycentric transform of a shifted subsequence determined by the indices \(m + i( m,r ) + d( m,r )\), \( \dots \), \( m + i\,( m,r ) + t( m,r ) + d( m,r )\). In certain cases shifting results in improved approximation of a transformed quotient to the limit or formal limit of interest. Naturally many \(a( m,r )\)-systems satisfy the required orthogonality condition. Those chosen by the authors are derived by taking \(d( m,1 ) = 1\), \[ a( m,1 \| k ) = ( -1 )^{k} / \{ s( m + i\,( m,r ) + k+1 ) - s( m + i\,( m,r ) + k ) \} \] for \(k = 0,1\), by setting \[ q( a \| m,r | E ) =\{\sum a( m,r | k ) E^{k} |0 \leq k \leq t( m,r ) \} \] and by making use of a recursion of the form \[ q( a \| m,r+1 | E ) =p( a \| m,r+1 | E ) q( a \| m,r | E) \] where \(p\) has the form \[ p(a \| m,r | E ) =\{\sum p\,( a \| m,r | k ) E^{k} |0 \leq k \leq t\,(m,r+1 )-t( m,r ) \}. \] For the special case in which the \(s\)-sequence is that (a) of the successive partial sums of a generalised or extended hypergeometric series, closed forms of the coefficients \(p( a \| m,r | k )\) are obtained and in particular a connection with transformations given by [\textit{J. Cizek, J. Zamastil} and \textit{L. Skala}, J. Math. Phys. 44, No.~3, 962--968 (2003; Zbl 1061.81028)], in [\textit{E. J. Weniger}, J. Math. Phys. 45, No.~3, 1209--1246 (2004; Zbl 1070.81056)] and \textit{S. Lewanowicz} and \textit{S. Paszkowski} [Math. Comput. 64, No.~210, 691--713 (1995; Zbl 0826.33004)] is established. Closed expressions also derive from certain series whose terms involve functions satisfying homogeneous three term recursions. In this way the theory of the paper is connected with that of transformations (b) of Fourier series given by \textit{C. Oleksy} [Comput. Phys. Commun. 96, No.~1, 17--26 (1996; Zbl 0921.65096)] and (c) of orthogonal polynomials given by \textit{S. Paszkowski} [Numer. Algorithms 47, No.~1, 35--62 (2008; Zbl 1134.65004)] and \textit{H. H. H. Homeier} [pp. 47--50 of R. Gruber, M. Tomassini (Eds.), Proceedings of the joint EPS - APS International Conference on Physics Computing. Physics Computing 94, European Physical Society, Boite Postale 69, CH 1213 Petit - Lancy, Geneva, Switzerland]. The described method requires the derivation of values of the coefficients \(p( a \| m,r | k )\). In the suggested implementation of the method the coefficients are treated as expressions obtained by the use of pencil and paper manipulations or the use of a computer symbol manipulation language; subsequently appropriate numerical values of \( m,r \) an \(k\) are to be plugged into these expressions to yield \(p\)-values. With commendable honesty the authors remark that such manipulations are not required in the use of other methods. Naturally it is possible to work directly with values, but in this case the given comparisons of effectiveness will have to be reformulated and stability attended to. The exposition is advanced by a succession of numerical examples: precious little analysis is offered. The reviewer suggests in particular that a case of equality between results produced by use of of the method and of the \(\epsilon\)-algorithm is not just a feature of the example being offered but holds for a class of such examples.