[Paper Review] A Dual Number Approach for Numerical Calculation of derivatives and its use in the Spherical 4R Mechanism

F. Pe, R. Pe+2 more|arXiv (Cornell University)|Jan 1, 2013

Robotic Mechanisms and Dynamics23 references3 citations

TL;DR

This paper introduces a dual number-based method for efficiently computing both first and second derivatives of vector functions in a single step, leveraging nested dual number arithmetic. Implemented in Fortran, the approach enables accurate and automated differentiation in kinematic analysis, demonstrated through application to the spherical 4R mechanism with improved computational efficiency and derivative accuracy.

ABSTRACT

This paper proposes a methodology to calculate both the rst and second derivatives of a vector function of one variable in a single computation step. The method is based on the nested application of the dual number approach for rst order derivatives. It has been implemented in Fortran language, a module which contains the dual version of elementary functions as well as more complex functions, which are common in the

Motivation & Objective

• To develop a unified computational method for calculating both first and second derivatives of vector functions in one operation.
• To extend the dual number approach—commonly used for first-order derivatives—toward higher-order derivatives.
• To implement the method in Fortran with a library of dual-number versions of elementary and complex functions.
• To apply the method to the kinematic analysis of the spherical 4R mechanism, a complex spatial mechanism requiring high-order derivatives.
• To improve computational efficiency and accuracy in numerical derivative calculations for mechanism design and analysis.

Proposed method

• The method employs nested dual numbers, where each dual number carries both function value and derivative information up to second order.
• First derivatives are computed using standard dual number arithmetic; second derivatives are obtained by applying dual arithmetic recursively to the first derivative components.
• A custom Fortran module implements dual-number versions of elementary functions (e.g., sin, cos, exp) and composite functions used in kinematics.
• The vector function of interest is evaluated using dual-number arithmetic, yielding both first and second derivatives simultaneously.
• The approach avoids symbolic differentiation and reduces computational overhead compared to finite difference or algorithmic differentiation with separate passes.
• The method is validated through application to the spherical 4R mechanism, where high-order derivatives are essential for dynamic and sensitivity analysis.

Experimental results

Research questions

• RQ1 Can dual number arithmetic be extended to compute second derivatives of vector functions in a single computational pass?
• RQ2 How does the nested dual number approach compare in accuracy and efficiency to traditional finite difference or separate first/second derivative computations?
• RQ3 What is the practical impact of this method on the kinematic analysis of complex mechanisms like the spherical 4R?
• RQ4 Can a robust Fortran library of dual-number functions be effectively implemented for engineering applications?
• RQ5 To what extent does the method improve the reliability of derivative-based calculations in mechanism design?

Key findings

• The method successfully computes both first and second derivatives of vector functions in a single evaluation using nested dual number arithmetic.
• The implementation in Fortran demonstrates computational efficiency, avoiding the need for multiple function evaluations or symbolic manipulation.
• The dual-number library enables accurate derivative computation for elementary and composite functions commonly used in kinematics.
• The approach is validated on the spherical 4R mechanism, where high-order derivatives are critical for dynamic and sensitivity analysis.
• The method achieves higher accuracy than finite difference approximations and reduces computational overhead compared to sequential derivative evaluation.
• The results confirm the feasibility and effectiveness of the dual number approach for higher-order derivatives in mechanical system analysis.