HOW TO DERIVE FORCE-FIELD PARAMETERS BY GENETIC ALGORITHMS - MODELINGTRIPOD-MO(CO)(3) COMPOUNDS AS AN EXAMPLE

Force field parameters used to describe the conformation of coordinati on compounds involving transition metals are generally derived by a tr ial-and-error procedure, until a somehow satisfying agreement between the calculated and observed conformations of a few members of a class of related compounds is reached. It is shown in this paper that a more general and less biased alternative is available, applicable to many structures at a time. Genetic Algorithms will effectively optimize for ce field parameters in an automatic way, on the basis of a potentially exhaustive set of all the structural data available for a given class of compounds. The feasibility of this procedure has been demonstrated by the derivation of force field parameters describing the conformati onal behaviour of tripod-Mo(CO)(3) compounds [tripod RCH2C(CH2X)(CH2Y) (CH(2)Z), X,Y,Z = PR'R '')] by simultaneous optimization based on the structure of ten individual molecules. With the force field parameters relevant to the organic part of these compounds taken from MM2,the p arameters involving contributions from the Mo center were refined. The agreement between observed and calculated structures is characterized by an rms deviation of around 0.3 Angstrom for the ten structures con tained in the data base. To assess the validity of this approach, the conformational space of CH3C(CH2PPh2)(3)Mo(CO)(3) was explored exhaust ively, A contour diagram representing the relative energy of the molec ule with respect to the rotational positions of its phenyl groups was found to effectively reproduce the scatter of these conformational par ameters as earlier derived from an analysis of 82 relevant compounds. - As a further assessment, the conformational space of CH3C[CH2P(o-Tol )(2)](3)Mo(CO)(3), which was not included in the data base, has been a nalyzed. It is found that the structure corresponding to the global en ergy minimum corresponds to that observed in the crystal with an rms d eviation of only 0.3 a. The novel approach to problems of this type - Genetic Algorithms had not previously been applied in this context - t hus appears promising.