Algorithmic differentiation and hull-consistency enforcing using C++ template meta-programming

Abstract Algorithmic differentiation is a tool used in several branches of computational science, both conjunction with the interval calculus, and apart it. This paper presents ADHC library [14] developed by author, identified as package “na60” Numerical Algorithms journal. makes intensive use C++ template meta-programming, it has unique features. seems particularly useful for interval-related applications. The been some solvers, also including HIBA_USNE. describes library, presenting its features, focusing on new ones, added version 2.0; particular, we describe bounding subdifferentials non-smooth functions computing derivatives over various datatypes. Efficiency comparison respect to other packages presented. Then examples applications are given, involving HIBA_USNE standard benchmark problems solving nonlinear systems. A particular emphasis put related modern machine learning, but not limited them. Planned extensions possible directions future development outlined discussed.