In this study, we propose a new global procedure to perform optimization of semiempirical intermolecular potential parameters on the basis of a large reference database. To obtain transferable parameters, we used the original method proposed by Ungerer ͓Ungerer et al., J. Chem. Phys. 112, 5499 ͑2000͔͒, based on the minimization of a dimensionless error criterion. This method allows the simultaneous optimization of several parameters from a large set of reference data. However, the computational cost of such a method limits its application, because it implies the calculation of an important number of partial derivatives, calculated by finite differences between the results of several different simulations. In this work, we propose a new method to evaluate partial derivatives, in order to reduce the computing time and to obtain more consistent derivatives. This method is based on the analysis of statistical fluctuations during a single simulation. To predict equilibrium properties of olefins, we optimize the Lennard-Jones potential parameters of the unsaturated hydrocarbon groups using the anisotropic united atoms description. The resulting parameters are consistent with those previously determined for linear and branched alkanes. Test simulations have been performed at temperatures ranging from 150 to 510 K for several ␣-olefins ͑ethylene, propene, 1-butene, 1-pentene, 1-hexene, 1-octene͒, several ␤-olefins ͑trans-2-butene, cis-2-butene, trans-2-pentene͒, isobutene, and butadiene. Equilibrium properties are well predicted, and critical properties can be evaluated with a good accuracy, despite the fact that most of the results constitute pure predictions. It is concluded that the AUA potential, due to a relevant physical meaning, can be transferred to a large range of olefins with good success.

The anisotropic united atoms potential for linear alkanes proposed by Ungerer (J. Chem. Phys., 112, 5499, 2000), called AUA4, has been used to predict several equilibrium properties (vapour pressure, vaporisation enthalpies, and liquid densities) of alkanes by Gibbs ensemble Monte Carlo simulation. In order to extend the potential to branched alkanes, potential parameters for the CH group have been determined by optimisation on the basis of equilibrium properties of isobutane, keeping the same parameters as AUA4 for the CH 3 groups. The resulting CH parameters form a regular sequence with those previously determined for CH 3 and CH 2 groups, so that a physically consistent parameter set is obtained.Simulations have been performed at temperatures ranging from 450 to 800 K for long n-alkanes (C20, C25 and C30) and from 350 to 450 K for four heptane isomers (n-heptane, 2-methylhexane, 2,4-dimethylpentane and 2-ethylpentane). In order to achieve internal relaxation of long chains with a good efficiency, a specific Monte Carlo move was used in which a united atom is rotated around its nearest neighbours. Equilibrium properties of long chain alkanes are well predicted, and small differences between heptane isomers are represented with a good accuracy. It is concluded that the AUA4 potential shows an interesting degree of transferability.