Adifor: a Fortran System for Portable Automatic Differentiation

Automatic diierentiation provides the foundation for sensitivity analysis and subsequent design optimization of complex systems by reliably computing derivatives of large computer codes, with the potential of doing it many times faster compared to current approaches. This paper describes the ADIFOR (Automatic DIIerentiation of FORtran) system, a translator that augments Fortran programs with statements for the computation of derivatives. ADIFOR accepts arbitrary Fortran 77 code deening the computation of a function and writes portable Fortran 77 code for the computation of its derivatives. Our goal is to free the computational scientist from worrying about the accurate and eecient computation of derivatives, even for complicated \functions", thereby enabling him to concentrate on the more important issues of system model-ing and algorithm design. This paper gives an overview of the principles underlying the ADIFOR system, and comments on the power of automatic diierentiation for computing derivatives of implicitly-deened functions.