Old and new multidimensional convergence accelerators (Q581990)

The author studies some classical multimensional convergence accelerators and proves that a number of these multidimensional convergence accelerators are particular cases of a whole class of multidimensional accelerators and also that some new multidimensional convergence accelerators can be introduced. The common undelying principle is that they can be considered as multivariate Padé approximants either for a multivariate function that is different for different algorithms. Also, several new multidimensional convergence accelerators are given by using multivariate rational Hermite interpolants instead of multivariate Padé approximants.