Using automatic differentiation to compute derivatives for a quantum-chemical computer program

The ADIFOR 2.0 tool for Automatic Differentiation of Fortran programs has been used to generate analytic gradient code for all semiempirical SCF methods available in the MNDO97 program. The correctness and accuracy of the new code have been verified. Its performance has been compared with that of hand-coded analytic derivative routines and with numerical differentiation. From a quantum-chemical point of view, the major advance of this work is the development of previously unavailable analytic gradient code for the recently proposed OM1 and OM2 methods.