A chain rule for multivariate divided differences (Q1960208)

Divided differences of composite functions appear in formulas for parametric curve fitting based on polynomials and splines [see \textit{M. S. Floater}, BIT 45, No.~4, 679--694 (2005; Zbl 1095.65011), IMA J. Numer. Anal. 26, No.~1, 25--33 (2006; Zbl 1095.41006)], \textit{K. Mørken} and \textit{K. Scherer} [Math. Comput. 66, No.~217, 237--260 (1997; Zbl 0854.41001)]. The author shows that at least the inner chain rule has a natural extension to multivariate functions when the divided differences are defined over rectangular grids of points. When compared with derivative chain rules as Faa di Bruno's formula [see \textit{C. F. Faa di Bruno}, ``Note sur une nouvelle formule de calcul differentiel'', Q. J. Pure Appl. Math. 1, 359--360 (1857)], an advantage of the divided difference rules is that they can be applied to functions with little or no smoothness. In Section 2, the author recalls the inner chain rule for univariate functions [see \textit{M. S. Floater} and \textit{T. Lyche}, Math. Comput. 76, No.~258, 867--877 (2007; Zbl 1113.05011)]. Section 3 is devoted to a chain rule for multivariate inner functions while Section 5 contains a chain rule for both multivariate functions. In Section 5 a chain rule for derivatives is obtained.