Data structure and algorithms for fast automatic differentiation

Abstract

Abstract In this paper we discuss the data structure and algorithms for the direct application of generalized Leibnitz rules to the numerical computation of partial derivatives in forward mode. The proposed data structure provides constant time access to the partial derivatives, which accelerates the automatic differentiation computations. The interaction among elements of the data structure is explained by several numerical examples. The paper contains analysis of the developed data structure and algorithms. Copyright © 2003 John Wiley & Sons, Ltd.