Higher-Order Automatic Differentiation Using Symbolic Differential Algebra: Bridging the Gap between Algorithmic and Symbolic Differentiation
Higher-Order Automatic Differentiation Using Symbolic Differential Algebra: Bridging the Gap between Algorithmic and Symbolic Differentiation
In scientific computation, it is often necessary to calculate higher-order derivatives of a function. Currently, two primary methods for higher-order automatic differentiation exist: symbolic differentiation and algorithmic automatic differentiation (AD). Differential Algebra (DA) is a mathematical technique widely used in beam dynamics analysis and simulations of particle accelerators, and it also functions as an algorithmic automatic differentiation method. DA automatically computes the Taylor expansion of a function at a specific point up to a predetermined order and the derivatives can be easily extracted from the coefficients of the expansion. We have developed a Symbolic Differential Algebra (SDA) package that integrates algorithmic differentiation with symbolic computation to produce explicit expressions for higher-order derivatives using the computational techniques of algorithmic differentiation. Our code has been validated against existing DA and AD libraries. Moreover, we demonstrate that SDA not only facilitates the simplification of explicit expressions but also significantly accelerates the calculation of higher-order derivatives, compared to directly using AD.
He Zhang
数学物理学
He Zhang.Higher-Order Automatic Differentiation Using Symbolic Differential Algebra: Bridging the Gap between Algorithmic and Symbolic Differentiation[EB/OL].(2025-05-31)[2025-06-19].https://arxiv.org/abs/2506.00796.点此复制
评论