On condition numbers of the Shanks transformation (Q5949504): Difference between revisions

The condition number of a suitable real valued function of finitely many real variables is defined to be the 1-norm of its gradient. Using a preparatory lemma concerning the derivative of the quotient of two determinants which differ only in one column, an expression for the condition number of a determinantal formula featuring in convergence acceleration is derived. The expression is used to investigate the transformation of series of totally monotone and totally oscillating terms. The formula concerned is due to \textit{R. J. Schmidt} [Philos. Mag., VII. Ser. 32, 369-383 (1941; Zbl 0061.27109)]; it was later marketed by \textit{D. Shanks} [J. Math. Phys. 34, 1-42 (1955; Zbl 0067.28602)]. Having regard to the interplay of relationships in the academic world, there has been and possibly still is more to be gained by ascribing the formula to the second author; nevertheless it is due to Schmidt. Considerable experience had shown its use to be unstable in many cases; in attempting to redraft it in a more stable form the epsilon algorithm was discovered.