Automatic differentiation of large sparse systems

DOI10.1016/0165-1889(90)90023-AMaRDI QIDQ2277762zbMATH OpenFDO

Authors L. C. W. Dixon, Z. Maany, M. Mohseninia

Publication date 1990

Published in Journal of Economic Dynamics and Control (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1016/0165-1889(90)90023-A

zbMATH Keywords

automatic differentiationconjugate gradient algorithmDixon problemDixon-Maany problemgradientHessianminimization algorithmsNumerical resultsPowell problemsymbolic computationtruncated Newton method

Mathematics Subject Classification ID

Numerical differentiation (65D25) Numerical mathematical programming methods (65K05) Symbolic computation and algebraic computation (68W30) Nonlinear programming (90C30)

Review text

The authors are concerned with the task of calculating the gradient \(\nabla F(x)\) and the Hessian \(\nabla^ 2F(x)\) of a function F(x), \(x\in {\mathbb{R}}^ n\), by exploiting the features of modern computer languages (for instance, ADA, Pascal-Sc) which allow the definition of operators and data structures. Following previous works by \textit{A. Griewank} [Mathematical programming, Proc. 13th Int. Symp.,Tokyo/Jap. 1988, Math. Appl., Jap. Ser. 6, 83-107 (1989; Zbl 0696.65015)] and \textit{L. B. Rall} [``Optimal implementation of differential arithmetic, Computer arithmetic. Scientific computation and programming language (1987; Zbl 0614.65001)] which consider symbolic computation for automatic differentiation, here five new algebras are introduced and discussed. The case of sparsity on the function F(x) is taken into account. The proposed automatic differentiation is then coupled with well-known minimization algorithms, the truncated Newton method as proposed by \textit{R. S. Dembo} [Math. Program. Study 31, 43-71 (1987; Zbl 0635.90072)] and the conjugate gradient algorithm based on the Polak-Ribière formula. Numerical results obtained in solving the extended Dixon problem, the extended Powell problem with \(n=4,8,20,40,80\) and the Dixon-Maany problem with \(n=3000\), are reported. This demonstrates widely the effectiveness of the approach considered by the authors.

Recommendations

• On the Impact of Automatic Differentiation on the Relative Performance of Parallel Truncated Newton and Variable Metric Algorithms
• Publication:3201636
• Truncated Newton method for sparse unconstrained optimization using automatic differentiation
• scientific article; zbMATH DE number 778143
• scientific article; zbMATH DE number 140022

Cites work

• A test problem generator for large-scale unconstrained optimization
• Automatic differentiation of large sparse systems
• scientific article; zbMATH DE number 4110213 (Why is no real title available?)
• scientific article; zbMATH DE number 4001347 (Why is no real title available?)
• scientific article; zbMATH DE number 879764 (Why is no real title available?)
• The Arithmetic of Differentiation
• Truncated Newton method for sparse unconstrained optimization using automatic differentiation

Cited in

(9)

• Truncated Newton method for sparse unconstrained optimization using automatic differentiation
• Automatic differentiation of large sparse systems
• Combining Automatic Differentiation Methods for High-Dimensional Nonlinear Models
• scientific article; zbMATH DE number 5353265 (Why is no real title available?)
• scientific article; zbMATH DE number 140022 (Why is no real title available?)
• Efficient computation of gradients and Jacobians by dynamic exploitation of sparsity in automatic differentiation
• Cheap Newton steps for optimal control problems: automatic differentiation and Pantoja's algorithm
• Computing the sparsity pattern of Hessians using automatic differentiation
• On interval enclosures using slope arithmetic

This page was built for publication: Automatic differentiation of large sparse systems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2277762)